Professional Documents
Culture Documents
Literal Meaning
François Recanati
Institut Jean-Nicod, Paris
CAMBRIDGE UNIVERSITY PRESS
Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, São Paulo
Introduction 1
1 Two approaches to ‘what is said’ 5
1.1 The basic triad 5
1.2 Minimalism 7
1.3 Literal truth-conditions vs actual truth-conditions 8
1.4 A problem for Minimalism 10
1.5 The availability of what is said 13
1.6 The availability based approach 16
1.7 ‘Saying’ as a pragmatic notion 18
1.8 Availability vs Minimalism 20
2 Primary pragmatic processes 23
2.1 Enrichment, loosening and transfer 23
2.2 Rejecting the Gricean picture 27
2.3 Accessibility 30
2.4 Objections and responses 32
2.5 Interactive processing 34
2.6 The role of schemata 36
3 Relevance-theoretic objections 38
3.1 One or two systems? 38
3.2 Personal and sub-personal inferences 40
3.3 Implicature or enrichment? 44
3.4 Mutual adjustment of explicature and implicature 46
3.5 Implicated premisses 48
3.6 Personal-level inferences: occurrent vs dispositional 49
4 The Syncretic View 51
4.1 Four levels? 51
4.2 Semantics and pragmatics: the literalist picture 54
4.3 Semantic underdeterminacy 56
4.4 The minimal proposition as ‘common denominator’ 58
4.5 Interaction between saturation and optional pragmatic processes 61
4.6 Do we need the minimal proposition? 64
4.7 The reflexive proposition 65
v
vi Contents
5 Non-literal uses 68
5.1 Non-literal uses as non-minimal departures from literal meaning 68
5.2 Non-literal uses and secondary meaning 70
5.3 Non-minimal departures without secondariness 72
5.4 The transparency condition 74
5.5 Varieties of non-literal meaning 75
5.6 Internal vs external duality 78
5.7 Conclusion 81
6 From Literalism to Contextualism 83
6.1 Five positions 83
6.2 Indexicalism 86
6.3 Contextualism 90
6.4 Literalist responses to the contextualist challenge 92
6.5 Where Indexicalism and Contextualism meet 95
7 Indexicalism and the Binding Fallacy 98
7.1 Mandatory vs optional 98
7.2 Two criteria 100
7.3 The indexicalist challenge 103
7.4 Is the Binding Criterion reliable? 105
7.5 Variadic functions 107
7.6 The Binding Fallacy 109
7.7 Conclusion: the failure of Indexicalism 111
8 Circumstances of evaluation 115
8.1 Modality 115
8.2 Time and tense 118
8.3 Situations 121
8.4 Saturation or enrichment? 124
8.5 Sub-sentential circumstances 125
8.6 Conclusion 127
9 Contextualism: how far can we go? 131
9.1 The modulation of sense 131
9.2 The semantic relevance of modulation 133
9.3 Four approaches 136
9.4 Truth-conditional unstability: from Waismann’s ‘open texture’
to Searle’s ‘background’ 141
9.5 Ostensive definitions 144
9.6 Meaning Eliminativism 146
9.7 Conclusion 151
Conclusion 154
10.1 Alleged arguments against Contextualism 154
10.2 Remnants of Literalism 159
10.3 Availability, Minimalism, and the dispositional/occurrent contrast 162
Bibliography 166
Index 175
Acknowledgments
This book started its life as a series of lectures at the University of California,
Berkeley, in 1994. Those who attended my graduate seminar in philosophy
that year were so passionately involved in discussing the foundational issues I
had raised that we all retain wonderful memories of those weeks of continuing
debate. (Or at least, I do.) The most active debaters were, undoubtedly, Herman
Cappelen and Josh Dever on the student side, and John Searle and Stephen
Neale on the faculty side. I am greatly indebted to the four of them for those
valuable and exciting discussions.
The second major step was taken when Professor Kunihiko Imai, of
Gakushuin University, invited me to present my views on the seman-
tics/pragmatics interface during a special workshop which took place in Tokyo
on 30 September 2001. For that workshop I prepared a long talk which, I soon
realized, could easily be expanded into a book. A couple of years earlier I
had contracted with Cambridge University Press for a book on literal meaning.
(The original title was ‘Context and Content’, but Robert Stalnaker published
a collection of papers under that title in 1999, so I had to find something else.)
I decided to use the Tokyo presentation as the nucleus for that book. I am
grateful to Professor Imai for the invitation, and for the discussions which took
place during the workshop. I also benefited from insightful comments by Yuji
Nishiyama, Haruhiko Yamaguchi and Seiji Uchida.
When the book was well under way the department of philosophy of the Uni-
versity of Granada (Spain), in charge of the thirteenth Inter-University Work-
shop on Philosophy and Cognitive Science to be held in February 2003, decided
to invite me as main speaker and to organize the workshop around my work.
I was supposed to give three talks during the three days of the workshop. I chose
to devote the three of them to the Literalism/Contextualism controversy, which
is the topic of this book. This provided me with a welcome opportunity for test-
ing my new ideas; an opportunity for which I wish to thank Maria José Frápolli,
Esther Romero and Belén Soria, as well as the SEFA (Sociedad Española de
Filosofı́a Analı́tica) in cooperation with which the inter-university workshops
are organized.
vii
viii Acknowledgments
During the past ten years, I have had many occasions for discussing those
issues with the fifteen to twenty philosophers and linguists who regularly gather
in conferences on the semantics/pragmatics distinction, contribute to the same
issues of the same journals, and so on. For fear of forgetting someone, I will not
list them individually here, but I thank them collectively; they know who they
are. Two persons in that crowd deserve special thanks: Robyn Carston, who
provided detailed, chapter-by-chapter comments on a first version of the book;
and Jason Stanley, whose systematic defence of the positions I attack provided
a helpful and timely challenge. I am also grateful to my students and colleagues
in Paris for numerous discussions which shaped my thinking on those topics.
Around the middle of the twentieth century, there were two opposing camps
within the analytic philosophy of language. The first camp – i d e a l l a n -
g ua g e p h i l o s o p h y , as it was then called – was that of the pioneers, Frege,
Russell, Carnap, Tarski, and so on. They were, first and foremost, logicians
studying formal languages and, through them, ‘language’ in general. They
were not originally concerned with natural language, which they thought de-
fective in various ways;1 yet, in the 1960s, some of their disciples established
the relevance of their methods to the detailed study of natural language.2 Their
efforts gave rise to contemporary f o r m a l s e m a n t i c s , a very active disci-
pline whose stunning developments in the last quarter of the twentieth century
changed the face of linguistics.
The other camp was that of so-called o r d i n a r y l a n g ua g e p h i l o s o -
p h e r s , who thought important features of natural language were not revealed
but hidden by the logical approach initiated by Frege and Russell. They ad-
vocated a more descriptive approach and emphasized the pragmatic nature of
natural language as opposed to, say, the language of Principia Mathematica.
Their own work3 gave rise to contemporary pragmatics, a discipline which,
like formal semantics, developed successfully within linguistics in the past
forty years.
Central in the ideal language tradition had been the equation of, or at least
the close connection between, the meaning of a (declarative) sentence and its
truth-conditions. This truth-conditional approach to meaning is perpetuated,
to a large extent, in contemporary formal semantics. A language is viewed as
a system of rules or conventions, in virtue of which certain assemblages of
1 There are a few exceptions. The most important one is Hans Reichenbach, whose insightful
‘Analysis of conversational language’ was published as a chapter – the longest – in his Elements
of Symbolic Logic (Macmillan, 1947).
2 See Richard Montague, Formal Philosophy: Selected Papers (Yale University Press, 1974), and
Donald Davidson, Inquiries into Truth and Interpretation (Clarendon Press, 1984).
3 The most influential authors were Austin, Strawson, Grice and the later Wittgenstein. Grice is
a special case, for he had, as he once said, one foot in each of the two camps (Paul Grice,
‘Retrospective Epilogue’, in his Studies in the Way of Words (Harvard University Press, 1989),
p. 372).
1
2 Literal Meaning
4 See my ‘Pragmatics and Semantics’, in Larry Horn and Gregory Ward (eds.), Handbook of
Pragmatics (Blackwell, forthcoming).
Introduction 3
There is much that is correct in this description, but there also is some-
thing which I think must be rejected, namely the contrast between literal truth-
conditions and speaker’s meaning. That contrast commits us to Literalism, and
in this book I want to argue for Contextualism. According to Contextualism,
the contrast between what the speaker means and what she literally says is il-
lusory, and the notion of ‘what the sentence says’ incoherent. What is said (the
truth-conditional content of the utterance) is nothing but an aspect of speaker’s
meaning. That is not to deny that there is a legitimate contrast to be drawn
between what the speaker says and what he or she merely implies. Both, how-
ever, belong to the realm of ‘speaker’s meaning’ and are pragmatic through and
through.
I will not only criticize Literalism and argue for Contextualism in the follow-
ing chapters. I will discuss all sorts of intermediate positions corresponding to
views actually held in the current debate about the semantics/pragmatics inter-
face. Whether or not one accepts my arguments, I hope the survey of logical
space which I provide will be useful to those interested in the debate, and will
contribute to shaping it in the years to come.
1 Two approaches to ‘what is said’
5
6 Literal Meaning
sentence meaning
vs
what is said
!
speaker’s meaning
what is implicated
Essential to this interpretation is the claim that ‘what is said’, though constrained
by the meaning of the sentence, is not as tightly constrained as is traditionally
thought and, in particular, does not obey what I will refer to as the ‘minimalist’
constraint.
Two approaches to ‘what is said’ 7
1.2 Minimalism
As I said above, what distinguishes ‘what is said’ from the implicatures is the
fact that the former must be ‘closely related to the conventional meaning of the
words (the sentence) [one] has uttered’.2 However, this constraint can be con-
strued more or less strictly. What I call ‘Minimalism’ construes the constraint
very strictly; ‘what is said’, in the minimalist framework, departs from the con-
ventional meaning of the sentence (and incorporates contextual elements) only
when this is necessary to ‘complete’ the meaning of the sentence and make it
propositional. In other words, the distance between sentence meaning and what
is said is kept to a minimum (hence the name ‘Minimalism’).
The crucial notion here is that of ‘saturation’. Saturation is the process
whereby the meaning of the sentence is completed and made propositional
through the contextual assignment of semantic values to the constituents of
the sentence whose interpretation is context-dependent (and, possibly, through
the contextual provision of ‘unarticulated’ propositional constituents, if one as-
sumes, as some philosophers do, that such constituents are sometimes needed
to make the sentence fully propositional). This process takes place whenever
the meaning of the sentence includes something like a ‘slot’ requiring comple-
tion or a ‘free variable’ requiring contextual instantiation.3 Thus an indexical
sentence like ‘He is tall’ does not express a complete proposition unless a ref-
erent has been contextually assigned to the demonstrative pronoun ‘he’, which
acts like a free variable in need of contextual instantiation. Genitives provide
another well-known example: an utterance including the phrase ‘John’s book’
does not express a complete proposition unless a particular relation has been
identified as holding between the book and John. Nominal compounds work the
same way: ‘burglar nightmare’ means something like ‘a nightmare that bears
a certain relation R to burglars’, which relation must be contextually identi-
fied. Other well-known examples of saturation include parametric predicates
(‘small’, ‘on the left’), definite null instantiation (that is, the case where one
of the arguments in the semantic structure of a lexeme, typically a verb, is not
syntactically realized and must be contextually identified, as when someone
says ‘I heard’ or ‘I noticed’), and so on and so forth.
Whenever saturation is in order, appeal to the context is necessary for the
utterance to express a complete proposition: from a semantic point of view,
saturation is a mandatory contextual process. Other contextual processes –
for example, the inference process generating implicatures – are semantically
optional in the sense that the aspects of meaning they generate are dispens-
able; the utterance would still express a complete proposition without them.
According to Minimalism, those extra constituents of meaning which are not
necessary for propositionality are external to what is said. The only justifica-
tion for including some pragmatically determined constituent of meaning into
what is said (as opposed to what is merely conveyed) is the indispensability of
such a constituent – the fact that the utterance would not express a complete
proposition if the context did not provide such a constituent.
In all such cases, as we shall see, the minimalist constraint implies that what
the utterance literally says is not what intuitively seems to be said.
From a minimalist point of view, the first sentence, ‘I’ve had breakfast’,
expresses the proposition that S (the speaker) has had breakfast before t* (the
time of utterance). Strictly speaking this proposition would be true if the speaker
had had breakfast twenty years ago and never since. This is clearly not what
the speaker means (when she answers the question ‘Do you want something to
eat?’ and replies ‘I’ve had breakfast’); she means something much more specific,
namely that she’s had breakfast on that very day (that is, the day which includes
t*). This aspect of speaker’s meaning, however, has to be construed as external to
what is said and as being merely conveyed, in the same way in which the utterer
of ‘I am French’ implies, but does not say, that he is a good cook. That is so
because the ‘minimal’ interpretation, to the effect that the speaker’s life was not
entirely breakfastless, is sufficient to make the utterance propositional. Nothing
in the sentence itself forces us to bring in the implicit reference to a particular
time span. Indeed we can easily imagine contexts in which a speaker would
use the same sentence to assert the minimal proposition and nothing more.4
The same thing holds even more clearly for the second example. Kent Bach,
to whom it is due, imagines a child crying because of a minor cut and her mother
uttering (2) in response. What is meant is: ‘You’re not going to die from that
cut.’ But literally the utterance expresses the proposition that the kid will not
die tout court – as if he or she were immortal. The extra element contextually
4 Dan Sperber and Deirdre Wilson, Relevance: Communication and Cognition (Blackwell, 1986),
pp. 189–90. For an alternative analysis of that example, see my ‘Pragmatics of What is Said’, in
Mind and Language 4 (1989), pp. 305–6, and §6.2 below.
Two approaches to ‘what is said’ 9
provided (the implicit reference to the cut) does not correspond to anything
in the sentence itself; nor is it an unarticulated constituent whose contextual
provision is necessary to make the utterance fully propositional. Again, we
can easily imagine a context in which the same sentence would be used to
communicate the minimal proposition and nothing more.5
What about (3)? John Perry and many others after him have argued as fol-
lows.6 Even though nothing in the sentence ‘It’s raining’ stands for a place,
nevertheless it does not express a complete proposition unless a place is con-
textually provided. The verb ‘to rain’, Perry says, denotes a dyadic relation – a
relation between times and places. In a given place, it doesn’t just rain or not,
it rains at some times while not raining at others; similarly, at a given time,
it rains in some places while not raining in others. To evaluate a statement of
rain as true or false, Perry says, we need both a time and a place. Since the
statement ‘It is raining’ explicitly gives us only the two-place relation (supplied
by the verb) and the temporal argument (indexically supplied by the present
tense), the relevant locational argument must be contextually supplied for the
utterance to express a complete proposition. If Perry is right, the contextual
provision of the place concerned by the rain is an instance of saturation, like
the assignment of a contextual value to the present tense: both the place and
the time are constituents of what is said, even though, unlike the time, the place
remains unarticulated in surface syntax.
But is Perry right? If really the contextual provision of a place was manda-
tory, hence an instance of saturation, every token of ‘It’s raining’ would be
unevaluable unless a place were contextually specified. Yet I have no difficulty
imagining a counterexample, that is, a context in which ‘It is raining’ is evaluable
even though no particular place is contextually singled out. In ‘Unarticulated
Constituents’ I depicted an imaginary situation in which
rain has become extremely rare and important, and rain detectors have been disposed
all over the territory (whatever the territory – possibly the whole Earth). In the imagined
scenario, each detector triggers an alarm bell in the Monitoring Room when it detects
rain. There is a single bell; the location of the triggering detector is indicated by a light
on a board in the Monitoring Room. After weeks of total drought, the bell eventually
rings in the Monitoring Room. Hearing it, the weatherman on duty in the adjacent room
shouts: ‘It’s raining!’ His utterance is true, iff it is raining (at the time of utterance) in
some place or other.7
The fact that one can imagine an utterance of ‘It’s raining’ that is true iff it is
raining (at the time of utterance) in some place or other arguably establishes
5 Kent Bach, ‘Conversational Impliciture’, in Mind and Language 9 (1994), p. 134. For an alter-
native analysis of that example (in terms of domain restriction), see below § 6.2.
6 John Perry, ‘Thought Without Representation’ (1986), reprinted (with a postscript) in his collec-
tion The Problem of the Essential Indexical and Other Essays (Oxford University Press, 1993),
205–25.
7 Recanati, ‘Unarticulated Constituents’, p. 317.
10 Literal Meaning
the pragmatic nature of the felt necessity to single out a particular place, in
the contexts in which such a necessity is indeed felt. When a particular place
is contextually provided as relevant to the evaluation of the utterance, it is
for pragmatic reasons, not because it is linguistically required. (Again, if it
were linguistically required, in virtue of semantic properties of the sentence
type, it would be required in every context.) If this is right, then the contextual
provision of a place is not an instance of saturation after all: it’s not something
that’s mandatory. It follows (by minimalist standards) that the place is not a
constituent of what is strictly and literally said: when I say ‘It is raining’ (rather
than something more specific like ‘It’s raining in Paris’ or ‘It’s raining here’),
what I literally say is true iff it’s raining somewhere or other.8 That is obviously
not what I mean, since what I mean involves a particular place. Appearances
notwithstanding, the situation is similar to the case of ‘I’ve had breakfast’,
where a restricted time interval is contextually provided for pragmatic reasons,
without being linguistically mandated.
Examples (4) and (5) are amenable to the same sort of treatment. According
to standard Russellian analysis, a definite description conveys an implication of
uniqueness: hence ‘The table is covered with books’ is true iff there is one and
only one table and it is covered with books. To make sense of this, we need either
to focus on a restricted situation in which there is indeed a single table, or to
expand the predicate ‘table’ and enrich it into, say, ‘table of the living-room’ in
order to satisfy the uniqueness constraint. Either way, it is arguable that the form
of enrichment through which we make sense of the utterance is not linguistically
mandated: it is only pragmatically required. If we don’t enrich, what we get is an
already complete proposition (albeit one that is pretty absurd): the proposition
that the only existing table is covered with books. Similarly with example (5):
without enrichment the utterance expresses a proposition that is true iff every
existing person went to Paris. Such a proposition is unlikely to be true, but
that does not make it incomplete. On this view the enrichment process through
which, in context, we reach the proposition actually communicated (to the effect
that everybody in such and such group went to Paris) is not linguistically but
pragmatically required; hence it is not an instance of saturation, but an optional
process of ‘free enrichment’. It follows that, in those examples as much as in the
previous ones, the proposition literally expressed is different from, and more
general than, the proposition actually communicated.
8 See Emma Borg, ‘Saying What You Mean: Unarticulated Constituents and Communication’
(forthcoming) for a defence of that claim.
Two approaches to ‘what is said’ 11
9 See Larry Horn, The Natural History of Negation (Chicago University Press, 1989), pp. 205–16.
10 As Grice puts it in one of his early papers, ‘one should not make a weaker statement rather
than a stronger one unless there is a good reason for so doing’ (Paul Grice, ‘The Causal Theory
of Perception’, in Proceedings of the Aristotelian Society, Supplementary Volume 35 (1961),
p. 132). Since the statement that John has (at least) three children is weaker than the statement
that John has n children (for n > 3), the maxim is obeyed only if John has no more than three
children. (If John has more than three children, the statement that he has three is too weak and
violates the maxim.) The statement ‘John has three children’ therefore implicates that John has
no more than three children, in virtue of the presumption that the maxim is obeyed.
12 Literal Meaning
what is communicated
CONSCIOUS
connection between them. On the other hand, there are the cases illustrated by
(1)–(6). Given his willingness to treat certain aspects of the intuitive meaning
of (1)–(6) as conversational implicatures external to what is literally said, the
minimalist must explain why those implicatures, unlike the prototypical cases
(for instance the French/cook example), do not have the property of conscious
‘availability’.
The only explanation I have come across in the literature makes use of Grice’s
distinction between ‘generalized’ and ‘particularized’ conversational implica-
tures, that is, between implicatures which arise ‘by default’, without any partic-
ular context or special scenario being necessary, and those which require such
specific contexts. In contrast with the latter, the former are ‘hard to distinguish
from the semantic content of linguistic expressions, because such implicatures
[are] routinely associated with linguistic expressions in all ordinary contexts’.11
Generalized implicatures are unconsciously and automatically generated and
interpreted. They belong to the ‘micropragmatic’ rather than to the ‘macroprag-
matic’ level, in Robin Campbell’s typology:
11 Stephen Levinson, Pragmatics (Cambridge University Press, 1983), p. 127. See also his Pre-
sumptive Meanings: The Theory of Generalized Conversational Implicature (MIT Press, 2000).
12 Robin Campbell, ‘Language Acquisition, Psychological Dualism and the Definition of Pragmat-
ics’, in Herman Parret, Marina Sbisà and Jeff Verschueren (eds.), Possibilities and Limitations
of Pragmatics (Benjamins, 1981), p. 101.
Two approaches to ‘what is said’ 13
But there are problems with this explanation. According to Horn, the gener-
alized nature of an implicature does not entail its conscious unavailability – its
‘cryptic’ character.13 In other words, it is possible for an implicature to be both
‘generalized’ and intuitively accessible as an implicature distinct from what is
said. Thus Horn insists that the generalized scalar implicature from ‘some’ to
‘not all’ is consciously available (in contrast to that from ‘three’ to ‘exactly
three’). A speaker saying ‘Some students came to the meeting’ normally im-
plies that not all students came, and when this is so there is no tendency on
the part of the interpreter to conflate the implicature with what is said. This is
actually debatable, for the ‘implicature’ at issue can arise at sub-sentential level
(for example, ‘He believes some students came’), and in such cases there are
reasons to doubt that the availability condition is satisfied. Be that as it may, the
‘generalization’ of an implicature does not seem to be necessary for its uncon-
scious character. Many particularized ‘bridging’ inferences are automatic and
unconscious. To take an example from Robyn Carston, ‘He went to the cliff
and jumped’ is readily interpreted as saying that the person referred to jumped
over the cliff, even though this is only contextually suggested.
13 Larry Horn, ‘The Said and the Unsaid’, in Chris Barker and David Dowty (eds.), SALT 2: Pro-
ceedings of the Second Conference on Semantics and Linguistic Theory (Ohio State University
Working Papers in Linguistics 40, 1992), 163–92.
14 See ‘The Pragmatics of What is Said’, already cited; Direct Reference: From Language to
Thought (Blackwell, 1993), pp. 233–74; and ‘What is Said’, in Synthèse 128 (2001), 75–91.
14 Literal Meaning
SUB-PERSONAL LEVEL
sentence meaning contextual ingredients
of what is said
Figure 1.2 An alternative approach
I therefore tend to agree with Bach’s criticism of the experiments through which
Gibbs and Moise attempted to support the availability based approach:17
[They] thought they could get their data about what is said, and thereby test the validity
of Recanati’s Availability Principle, by asking people what is said by a given utterance,
or by asking them whether something that is conveyed by a given utterance is implicated
or merely said. Evidently they assume that what people say about what is said is strongly
indicative of what is said. In fact, what it is indicative of is how people apply the phrase
‘what is said’ . . . It tells us little about what is said, much less about the cognitive
processes whereby people understand utterances.18
However, Bach himself uses what he calls the ‘IQ test’ to determine what
is said, that is, he ties what is said to indirect speech reports of what is said.19
I find this procedure most objectionable, and that is not what I mean when
I claim that what is said should be individuated according to the intuitions
of normal interpreters. Thus I strongly disagree with Cappelen and Lepore’s
surprising statement:
We ourselves don’t see how to elicit intuitions about what-is-said by an utterance of a
sentence without appealing to intuitions about the accuracy of indirect reports of the
form ‘He said that . . .’ or ‘What he said is that . . .’ or even ‘What was said is that . . .’20
availability based approach. The speaker’s intuitions concerning what is said need not involve
the very notion of what is said.
17 Raymond Gibbs and Jessica Moise, ‘Pragmatics in Understanding What is Said’, in Cognition
62 (1997), 51–74.
18 Kent Bach, ‘Seemingly Semantic Intuitions’, in Joseph Keim Campbell, Michael O’Rourke and
David Schier (eds.), Meaning and Truth (Seven Bridges Press, 2002), p. 27.
19 ‘IQ’ means i n d i r e c t q u o tat i o n . On the IQ test, see Bach’s papers ‘Semantic Slack’, in
Savas Tsohatzidis (ed.), Foundations of Speech Act Theory (Routledge, 1994), 267–91, ‘The
Myth of Conventional Implicature’, in Linguistics and Philosophy 22 (1999), 327–66, and ‘You
Don’t Say?’, in Synthèse 128 (2001), 15–44.
20 Herman Cappelen and Ernie Lepore, ‘On an Alleged Connection Between Indirect Speech and
the Theory of Meaning’, in Mind and Language 12 (1997), p. 280.
21 For an implicit use of that procedure, see Saul Kripke, Naming and Necessity (Blackwell, 1980),
p. 12 .
16 Literal Meaning
or false, but they were also given the option of leaving the matter open in case they
couldn’t make up their minds.22
26 As I pointed out in footnote 3, p. 7, that is true even when saturation consists in providing a
so-called ‘unarticulated constituent’.
Two approaches to ‘what is said’ 19
acknowledging that fact. According to Grice, when I say ‘John is a fine friend’
in the mentioned situation, I do not really say that John is a fine friend – I
pretend to be saying it. The pragmatic construal of saying forces Grice to draw
a distinction between ‘saying’ and ‘making as if to say’.
As far as I am concerned, I find Grice’s distinction (between genuine saying
and making as if to say) perfectly legitimate, but I can understand the worries of
those who feel that the notion of ‘saying’ he uses is too much on the pragmatic,
illocutionary side.27 We certainly need a notion of ‘what is said’ which captures
the objective content of an utterance irrespective of its pragmatic force as a
serious assertion or as an ironical utterance. Still, I find the objection superfi-
cial, for it is quite easy actually to construct the desired notion within Grice’s
own framework. Grice uses ‘say’ in a strict sense. In that sense whatever is said
must be meant. But we can easily define a broader sense for ‘say’:
S says that p, in the broad sense, iff he either says that p (in the strict sense) or makes as
if to say that p (again, in the strict sense of ‘say’).
I will henceforth use ‘say’ in that broad sense, which remains within the confines
of the pragmatic construal.
Another objection to the pragmatic construal focuses on the loss of objectivity
that allegedly goes with it. What is said is objective in the sense that it is possible
both for the speaker to make a mistake and say something other than what
he means, and for the hearer to misunderstand what the speaker is saying.
Those mistakes are possible, the objector will argue, because what is said is an
objective property of the sentence (in context). But on the pragmatic construal, it
is not clear that this objectivity can be captured. Imagine the following situation:
the speaker wants to say that Paul is tall, and, mistaking Tim for Paul, says ‘He is
tall’ while pointing to Tim. The speaker thus inadvertently says that Tim is tall.
Now imagine that the hearer also mistakes Tim for Paul. Thanks to this lucky
mistake, he grasps what the speaker means, thinking that this is what he has
said. The speaker and the hearer therefore converge on a certain interpretation,
which is not objectively what was said, but which they both (mistakenly) think
is what was said. How, in the framework I have sketched, will it be possible to
dissociate what is actually said from the protagonists’ mistaken apprehension
of what is said? Have we not equated what is said with their understanding of
what is said?
We have not. We have equated what is said with what a normal interpreter
would understand as being said, in the context at hand. A normal interpreter
knows which sentence was uttered, knows the meaning of that sentence, knows
27 ‘The verb “say”, as Grice uses it, does not mark a (locutionary) level distinct from that marked
by such illocutionary verbs as “state” and “tell”, but rather functions as a generic illocutionary
verb’ (Bach, ‘You Don’t Say?’, p. 41).
20 Literal Meaning
the relevant contextual facts (who is being pointed to, and so on).28 Ordinary
users of the language are normal interpreters, in most situations. They know
the relevant facts and have the relevant abilities. But there are situations (as in
the above example) where the actual users make mistakes and are not normal
interpreters. In such situations their interpretations do not fix what is said.
To determine what is said, we need to look at the interpretation that a normal
interpreter would give. This is objective enough, yet remains within the confines
of the pragmatic construal.
28 This is all tacit knowledge, not the sort of ‘conscious awareness’ I talk about in connection with
secondary pragmatic processes.
Two approaches to ‘what is said’ 21
is said. If we maintain that those ingredients are indeed ‘optional’ rather than
necessary for propositionality, this implies that we must give up the minimalist
criterion according to which the context contributes to what is said only when
this is necessary for some proposition to be expressed.
According to the view we arrive at, truth-conditional interpretation is prag-
matic to a large extent. Various pragmatic processes come into play in the very
determination of what is said; not merely saturation – the contextual assignment
of values to indexicals and free variables in the logical form of the utterance – but
also free enrichment and other processes which are not linguistically triggered
but are pragmatic through and through. Figure 1.3 summarizes the contrast be-
tween the two conceptions (Minimalism, and the availability based approach).
According to the availability based approach, the crucial distinction is not
between mandatory and optional contextual processes, but between those that
are ‘primary’ and those that are ‘secondary’. Primary pragmatic processes in-
clude not only saturation, but also ‘optional’ processes such as free enrichment.
Independent evidence for their inclusion in this category is provided by the
fact that, in general, the notion of ‘what is said’ we need to capture the input
to secondary, inferential processes already incorporates contextual elements of
the optional variety. Consider examples (1)–(6) once again. In each case we
may suppose that the speaker implies various things by saying what she does.
Thus, by saying that she’s had breakfast, the speaker implies that she is not
hungry and does not want to be fed. By saying that the child is not going to
die, the mother implies that the cut is not serious; and so forth. Now those
implicatures can be worked out only if the speaker is recognized as expressing
the (non-minimal) proposition that she’s had breakfast that morning, or that the
child won’t die from that cut. Clearly, if the speaker had had breakfast twenty
22 Literal Meaning
years ago (rather than that very morning), nothing would follow concerning
the speaker’s present state of hunger and her willingness or unwillingness to
eat something. The implicature could not be derived, if what the speaker says
was not given the richer, temporally restricted interpretation. If therefore we
accept the Gricean picture, according to which ‘what is said’ serves as input to
the secondary process of implicature-generation, we must, pace Grice himself,
acknowledge the non-minimal character of what is said. This provides some
support to the availability based approach, as against Minimalism.
2 Primary pragmatic processes
1 On the distinction between ‘pre-semantic’ and ‘post-semantic’ pragmatics, see Levinson, Pre-
sumptive Meanings, p. 187, and Ken Taylor, ‘Sex, Breakfast, and Descriptus Interruptus’, in
Synthèse 128 (2001), pp. 48–9.
23
24 Literal Meaning
conjunct; yet this is not explicitly articulated in the sentence. Insofar as the
bridging inference affects the intuitive truth-conditions of the utterance, it does
so as a result of free enrichment.2
In typical cases free enrichment consists in making the interpretation of
some expression in the sentence contextually more specific. This process has
sometimes been described in the literature as ‘specifization’. For example the
mass term ‘rabbit’ will be preferentially interpreted as meaning rabbit fur in
the context of ‘He wears rabbit’ and as meaning rabbit meat in the context of
‘He eats rabbit.’3 This not a matter of selecting a particular value in a finite set;
with a little imagination, one can think of dozens of possible interpretations
for ‘rabbit’ by manipulating the stipulated context of utterance; and there is
no limit to the number of interpretations one can imagine in such a way. Nor
can the process of specifization be construed as linguistically mandated, that
is, as involving a hidden variable. Were it linguistically mandated (bottom up),
it would be mandatory, but it is not: in some contexts the mass term ‘rabbit’
means nothing more than r a b b i t s t u f f (‘after the accident, there was rabbit
all over the highway’).
Can free enrichment be equated with specifization, or are there instances of
free enrichment that are not cases of specifization? The provision of (optional)
unarticulated constituents is supposed to be a case of free enrichment in which it
2 The term ‘bridging inference’ was originally introduced by Herb Clark, a pioneer of pragmatic
studies, in the seventies (see e.g. ‘Bridging’, in Peter Johnson-Laird and John Wason (eds.),
Thinking: Readings in Cognitive Science (Cambridge University Press, 1977), 411–20). Example
(1) is discussed by Robyn Carston in ‘Implicature, Explicature, and Truth-Theoretic Semantics’,
in Ruth Kempson (ed.), Mental Representations: The Interface between Language and Reality
(Cambridge University Press, 1988), 155–81.
3 This example is discussed in Geoff Nunberg and Annie Zaenen, ‘Systematic Polysemy in Lex-
icology and Lexicography’, in Hannu Tommola, Krista Varantola, Tarja Tolonen and Jürgen
Schopp (eds.), Proceedings of Euralex 2 (University of Tampere, 1992), 387–98. A number
of similar examples are discussed in the cognitive science literature on ‘concept combination’,
which parallels the semantics and pragmatics literature. Concept combination is said to require
not only specifization, that is, the addition of features, but also feature cancellation or loosening
(as in ‘stone lion’ or ‘fake gun’). See Bradley Franks, ‘Sense Generation: A “Quasi-Classical”
Approach to Concepts and Concept Combination’, in Cognitive Science 19 (1995), 441–505.
See also Richard Gerrig and Gregory Murphy, ‘Contextual Influences on the Comprehension of
Complex Concepts’, in Language and Cognitive Processes, 7 (1992), 205–30; Thomas Goschke
and Dirk Koppelberg, ‘Connectionist Representations, Semantic Compositionality, and the In-
stability of Concept Structure’, in Psychological Research 52 (1990), 253–70; Gregory Murphy,
‘Noun Phrase Interpretation and Conceptual Combination’, in Journal of Memory and Language
29 (1990), 259–88, and ‘The Comprehension of Complex Concepts’, in Cognitive Science 12
(1988), 529–62; Douglas Medin and Edward Shoben, ‘Context and Structure in Conceptual Com-
bination’, in Cognitive Psychology 20 (1988), 158–90; Jim Hampton, ‘Inheritance of Attributes
in Natural Concept Conjunctions’, in Memory and Cognition 15 (1987), 55–71. Further refer-
ences can be found in those papers. For relevant discussions, see also Paula Schwanenflugel (ed.),
The Psychology of Word Meanings (Lawrence Erlbaum Associates, 1991), and Raymond Gibbs,
The Poetics of Mind: Figurative Thought, Language, and Understanding (Cambridge University
Press, 1994).
Primary pragmatic processes 25
is not the interpretation of some expression in the sentence that is enriched, but
more globally the interpretation of the sentence. In most cases, however, what
can be done in terms of unarticulated constituents can also be done in terms of
specifization. We can construe the implicit instrument in the second conjunct of
(1) either as an unarticulated constituent (corresponding to the implicit prepo-
sitional phrase ‘with the key’), or as an aspect of the interpretation of the predi-
cate ‘open’ resulting from specifization (the concept contextually expressed by
‘open’ being the specific, ad hoc concept o p e n w i t h k e y , rather than the
generic concept o p e n ).4 The same options are presumably available for deal-
ing with the ‘rabbit fur/meat’ example. In such cases, I will assume that there is a
single form of free enrichment, which can be handled in different frameworks –
either in terms of specifization (ad hoc concepts) or in terms of unarticulated
constituents. Which framework we choose to handle such cases depends upon
extraneous considerations. (For example, if we want to preserve the principle
of compositionality, we’d better opt for the specifization view which spares us
the postulation of syntactically unarticulated constituents.) Still, there is a type
of case for which I think we need the notion of unarticulated constituent and
cannot make do with specifization and ad hoc concepts: whenever the alleged
unarticulated constituent is the intended ‘circumstance of evaluation’, we can’t
deal with it in terms of specifization or ad hoc concept. That type of case will
be discussed in chapter 8.
Another issue regarding enrichment is whether or not it can be described
as ‘strengthening’ or logical enrichment, as I suggested in Direct Reference
(p. 261). A predicate has conditions of application, and strengthening consists
in restricting the application of a predicate by contextually providing further
conditions that are not linguistically encoded. Thus ‘table’ has such and such
conditions of application packed into the concept ta b l e , and through contex-
tual strengthening the further condition i n t h e l i v i n g r o o m is provided,
which results in a restricted application. Thus construed enrichment can account
for the (so-called) contextual restriction of quantifiers and for the interpreta-
tion of (so-called) ‘incomplete’ definite descriptions. (‘All the books are on the
table’, where a particular set of books and a particular table are in question.)5
4 The notion of ‘ad hoc concept’, introduced by Larry Barsalou (see for example ‘Ad hoc Cate-
gories’, in Memory and Cognition 11 (1983), 211–27), now belongs to the toolkit of relevance
theory. See Dan Sperber and Deirdre Wilson, ‘The Mapping Between the Mental and the Public
Lexicon’, in Peter Carruthers and Jill Boucher (eds.), Language and Thought: Interdisciplinary
Themes (Cambridge University Press, 1998), 184–200; Robyn Carston, ‘Enrichment and Loosen-
ing: Complementary Processes in Deriving the Proposition Expressed?’, in Linguistische Berichte
8 (1997), 103–27, and Thoughts and Utterances: The Pragmatics of Explicit Communication
(Blackwell, 2002), chapter 5; Deirdre Wilson, and Dan Sperber, ‘Truthfulness and Relevance’,
in Mind 111 ( 2002), 583–632.
5 Stephen Neale (This, That and the Other (typescript), chapter 1) objects to my notion of strength-
ening: ‘It is sometimes said that enrichment in Sperber and Wilson’s sense involves strengthening
26 Literal Meaning
in that the post-enrichment statement entails the statement one might associate with the utter-
ance prior to enrichment. That this cannot be true in general is made clear by cases involving
(non-persistent) determiners like “every”, “no” and “some”. If, when discussing a dinner party
I attended last night, I say, “everyone drank a lot of wine”, I am saying only that everyone who
attended the dinner party drank a lot of wine, and this does not entail that everyone (unqualified)
drank a lot of wine.’
But I had already disposed of that objection in Direct Reference (pp. 262–3): ‘Strengthening, it
may be argued, can be understood as operating locally. For example, in the case of “Every boy
came”, we might say that it is the predicate “boy” that is strengthened into “boy in the class”,
rather than the proposition “Every boy came” into “Every boy in the class came”.’ (See also ‘The
Pragmatics of What is Said’, p. 307 n.)
6 Dan Sperber and Deirdre Wilson, ‘Loose Talk’, in Proceedings of the Aristotelian Society 86
(1986), 153–71; Carston, ‘Enrichment and Loosening’. Similar notions include ‘feature cancel-
lation’ (Jonathan Cohen, ‘Some Remarks on Grice’s Views About the Logical Particles of Nat-
ural Language’, in Yehoshua Bar-Hillel (ed.), Pragmatics of Natural Languages (Reidel, 1971),
50–68; Franks, ‘Sense Generation’), ‘concept broadening’ as opposed to ‘concept narrowing’
(Renate Bartsch, ‘The Structure of Word Meanings: Polysemy, Metaphor, Metonymy’, in Fred
Landman and Frank Veltman (eds.), Varieties of Formal Semantics (Foris Publications, 1984),
p. 32; see also Levinson, Presumptive Meanings), and ‘pragmatic generalization’ as opposed to
‘pragmatic specialization’ (Charles Ruhl, On Monosemy: A Study in Linguistic Semantics (State
University of New York Press, 1989)). Paul Ziff speaks of ‘augmenting’ and ‘diminishing’ senses
(‘What is Said’, in Donald Davidson and Gilbert Harman (eds.), Semantics of Natural Language
(Reidel, 1972), p. 719).
7 See Geoffrey Nunberg, ‘The Non-Uniqueness of Semantic Solutions: Polysemy’, in Linguistics
and Philosophy 3 (1979), 143–84, and ‘Transfers of Meaning’, in Journal of Semantics 12 (1995),
109–32.
Primary pragmatic processes 27
been accessed. This seems to support the Gricean picture, according to which we
process the literal interpretation first, and move on to the derived interpretation
only when this is required to make sense of the speaker’s utterance.
This argument is fallacious.8 I admit that the literal interpretation must come
first, insofar as the derived interpretation is derived from it through enrichment,
loosening or transfer. But I want to resist the conclusion that the literal inter-
pretation must be ‘processed’ first, in the sense that is relevant to the debate.
Or, to put it another way: what I am rejecting is not the claim that the literal
interpretation of the constituent is accessed before the derived interpretation –
that I take to be obvious – but the claim that a similar priority holds at the level
of the complete sentence; that is, I reject the claim that the process of seman-
tic composition begins by paying attention only to literal semantic values, and
turns to derived values only after the literal semantic value of the whole (the
proposition literally expressed) has been computed. It is this picture which I
think is unwarranted.
If I am right, the asymmetric dependence of derived meaning upon literal
meaning does not rule out an account according to which literal meaning and
derived meaning are on equal footing as far as semantic composition is con-
cerned. Thus we can imagine that the literal meaning and the derived meaning
of a given expression are processed in parallel, in constructing an interpretation
for the whole utterance. In the model I have in mind, the literal meaning of the
expression is accessed first and triggers the activation of associatively related
representations. That literal meaning is a natural candidate for the status of
semantic value, but there are others: some of the representations activated by
association contribute further candidates for the status of semantic value.9 All
candidates, whether literal or derived, are processed in parallel and compete.
When an interpretation which fits the broader context of discourse is found, it
is selected (that is, it undergoes semantic composition) and the other candidates
are suppressed.
On this view, derived meanings still proceed (associatively) from literal mean-
ings, which they indeed presuppose; but, although generated serially, they are
processed in parallel. The literal meaning has no compositional privilege over
derived meanings; they compete and it is possible for some derived mean-
ing to be retained (if it fits the broader context of discourse) while the literal
8 For a detailed critique, see my papers ‘Processing Models for Non-Literal Discourse’, in Roberto
Casati, Barry Smith and Graham White (eds.), Philosophy and the Cognitive Sciences, Proceed-
ings of the 16th International Wittgenstein Symposium (Hölder-Pichler-Tempsky, 1994), 277–90,
and ‘The Alleged Priority of Literal Interpretation’, in Cognitive Science 19 (1995), 207–232.
9 By ‘semantic value’ here I mean what the process of semantic composition operates on. My point
is that the semantic value contributed by an expression need not be the literal meaning of that
expression – it may be affected by pragmatic processes, including optional pragmatic processes.
(Later in this book I will talk of ‘pragmatic value’ instead of ‘semantic value’ to stress the fact
that such values result from the operation of such processes.)
Primary pragmatic processes 29
2.3 Accessibility
When someone talks of ‘wearing rabbit’, the literal meaning of the mass term
‘rabbit’ (namely r a b b i t s t u f f ) is accessed, but it has to compete with other
candidates for semantic value. The more specific representation r a b b i t f u r
is also activated since it is associatively connected to the representations en-
coded by both ‘rabbit’ and ‘wear’. As a result of this multiple activation, it is
possible for the representation r a b b i t f u r to be more active, in this context,
than the less specific representation r a b b i t s t u f f which is linguistically
encoded. Whatever we think of this particular example, it seems to me that
the following situation can obtain: an expression linguistically encodes a cer-
tain representation; that representation becomes active when the expression is
uttered, but another, associatively related representation is also activated (in
part – but in part only – through the encoded representation) and turns out to
be more active in that context than the original representation from which (in
part) it derives.13 If we assume that the candidate for semantic value which is
retained and undergoes semantic composition is that which is most accessible,
that is, that which corresponds to the most active representation when the inter-
pretation process stabilizes, we explain how a derived meaning resulting from
enrichment (or loosening, or transfer) can be selected as semantic value, in lieu
of the literal meaning from which it is derived.14
The same phenomenon can happen dynamically. Some representation may
be activated through its associative links to the representation linguistically
encoded and become more active than the latter as a result of the coming
into play of further linguistic material which raises its activation level. Before
considering such a case, let us look at a simpler example of accessibility shift
along the temporal dimension.
Consider sentence (2) from Direct Reference (p. 265):
(2) John was arrested by a policeman yesterday; he had just stolen a wallet.
In order to interpret the utterance one must assign a reference to the pronoun
‘he’ in the second sentence. The two persons who were mentioned in the first
sentence, namely John and the policeman, are obvious candidates, and we may
13 I assume that representations can be activated because they are being processed (or have been
recently), or because they receive activation from associatively related representations which
are themselves activated. Frequency of processing is another important factor (among several
others); it can be conceived of as lowering the activation threshold of a representation. See for
example Lawrence Barsalou and Dorrit Billman, ‘Systematicity and Semantic Ambiguity’, in
D. Gorfein (ed.), Resolving Semantic Ambiguity (Springer, 1989), 146–203.
14 The role of accessibility (or activation, or salience) has been stressed in a number of works. See
Mira Ariel, ‘Accessibility Theory: An Overview’, in Ted Sanders, Joost Schilperoord and Wilbert
Spooren (eds.), Text Representation: Linguistic and Psycholinguistic Aspects (John Benjamins,
2001), 29–87, and Rachel Giora, On Our Mind: Salience, Context, and Figurative Language
(Oxford University Press, forthcoming).
Primary pragmatic processes 31
15 On frames, see for example, Charles Fillmore and Beryl Atkins, ‘Toward a Frame-Based Lexicon:
The Semantics of r i s k and its Neighbors’, in Adrienne Lehrer and Eva Kittay (eds.), Frames,
Fields and Contrasts: New Essays in Semantic and Lexical Organization (Lawrence Erlbaum
Associates, 1992), 75–102. Other references are provided in footnote 29, p. 94.
32 Literal Meaning
(represented as) persons. In this way the representation of the ham sandwich
orderer gains some extra activation which makes him more accessible than the
ham sandwich, after the predicate has been processed.
But my account does not predict that the hearer would understand (ii) (‘The
ham sandwich is being eaten’) as being about the ham sandwich orderer. There
is something, in my account, which keeps the local process of transfer from
being triggered in a case like (ii). What keeps the local process of transfer from
being triggered in a case like (ii) is precisely what answers Bach’s legitimate
question: ‘How can the hearer go from the concept of ham sandwich to that
of ham-sandwich-orderer without first entertaining the absurd minimal [that is
literal] proposition?’ In my framework, the process of metonymical transfer
is ‘local’, not global, yet it is sensitive to the linguistic (and extralinguistic)
context in which the expression which receives the metonymical interpretation
occurs. In particular, it is sensitive to the meaning of the predicate expression;
this is why it is triggered in (i) but not in (ii).
For the literal proposition to be entertained, the literal semantic value of ‘(the)
ham sandwich’ must undergo semantic composition with the semantic value
of the predicate; but as soon as the semantic value of the predicate is accessed,
the metonymical interpretation of the subject-term becomes more accessible
than its literal interpretation, which is therefore prevented from going into the
proposition expressed. The activation of the concept which corresponds to
the literal interpretation of the predicate is both what triggers the process of
metonymical transfer in (i) and what prevents it from taking place in (ii). Yet
the transfer’s sensitivity to the interpretation of the other constituents in the
sentence does not mean that the transfer is a global process involving the com-
putation of the absurd, literal proposition. The interpreter does not go from the
concept of ham sandwich to that of ham-sandwich-orderer after having enter-
tained the absurd literal proposition; rather, it is because the interpreter goes
from the concept of ham sandwich to that of ham-sandwich-orderer (as a result
of an accessibility shift resulting from the interpretation of the predicate) that
he or she does not entertain the absurd literal proposition.
In this account, the metonymical transfer from ham sandwich to ham-
sandwich-orderer results from a shift in accessibility triggered by the inter-
pretation of another constituent in the sentence. The literal interpretation of the
subject-term was more accessible than the metonymical interpretation before
the predicate came into the picture, but the metonymical interpretation becomes
17 Ibid., p. 158.
34 Literal Meaning
A city is not the sort of thing that sleeps (in the normal sense of the term),
hence the overall interpretation of (3) is likely to involve some process of non-
literal interpretation. There are various possibilities, though. ‘The city’ can be
interpreted metonymically as standing for ‘the inhabitants of the city’ (transfer);
or ‘asleep’ can be interpreted metaphorically in the sense of ‘quiet and showing
little activity’ (loosening). Given the likelihood of a non-literal interpretation
for some constituent, if ‘the city’ literally applies to a city, ‘asleep’ will have to
be taken non-literally; conversely, if ‘asleep’ is literal, ‘the city’ will not be.
The same sort of trade-off applies to seemingly unproblematic examples such
as the following:
(4) I finished the book.
(5) John heard the piano.
If ‘finish’ has its standard sense in (4), its object (what is said to be finished)
must be a process, for only processes can start or finish in the standard sense. It
follows that ‘the book’ must be interpreted non-literally – it must be interpreted
as standing, not for a certain book, but for the process of, say, reading a certain
book. Conversely, if ‘the book’ is interpreted literally and stands for a certain
book, ‘finish’ must have a derived sense and mean something like finished
reading. Pustejovsky defends the former analysis, Langacker the latter.18
(on expansion and completion, see Kent Bach, Thought and Reference (Oxford University Press,
1987) and ‘Conversational Impliciture’).
19 A number of similar tests are offered by Geoff Nunberg in ‘Indexicality and Deixis’ (in Linguis-
tics and Philosophy 16 (1993), 1–43) and ‘Transfers of Meaning’. For example, Nunberg argues
that (a) and (b) must be analysed in fairly different ways, despite their superficial similarity:
(a) I am parked out back.
(b) This one <pointing to a car key> is parked out back.
Example (b) is a case of ‘deferred ostension’: an object (the key) is ‘demonstrated’, but the
ultimate ‘referent’, as opposed to the ‘demonstratum’, is another object suitably related to the
demonstratum, namely the car itself. In contrast, ‘I’ does not refer to the car in (a); it normally
refers to the speaker. It’s the predicate ‘parked out back’ that bears the onus of non-literality.
That the referent is the car rather than the (demonstrated) key in (b) is established by the fact that
semantic features of the referring expression such as number or gender do n o t correspond to
the key, but to the car; thus we would say ‘this one’ (in the singular) even if we were exhibiting
a whole set of keys for the car in question. Also, we can say things like:
(c) This <pointing to a car key>, which is an old Chevrolet, is parked out back.
The relative clause clearly qualifies the car, not the key. These tests provide evidence that the
referent is the car, even if it is the key which is ‘demonstrated’. When we turn to (a), the situation
is totally different. Gender and number correspond to the speaker, not to the car. We would not
say ‘We are parked out back’ if there was a single car owner and several cars (whereas we can
say ‘Those <pointing to a key> are parked out back’ if there is a single key and several cars).
And we can’t say ‘I, who am an old Chevrolet, am parked out back.’
36 Literal Meaning
The trade-off talked about in this section is a particular case of a more general
and fairly complex phenomenon: the search for coherence in interpretation. The
examples involve two constituents, A and B, each with (at least) two possible
interpretations (Ai and Aj, and Bi and Bj). Now there is a better ‘fit’ between
Ai and Bj or between Aj and Bi than there is between either Ai and Bi or Aj
and Bj, hence the ‘trade-off’ effect: the two global interpretations <Ai, Bj>
and <Aj, Bi> are favoured, in such a way that the choice of either Aj or Ai as
the correct interpretation for A ‘coerces’ a particular interpretation (Bi or Bj)
for B. The tendency to prefer coherent interpretations (with a high degree of fit
between the various semantic values) is what we must now try to account for.
But first, we must introduce the notion of a schema.20
Given the link we have established between coherence (or fit) and schemata,
the question we must answer becomes: why are schema-instantiating interpre-
tations more successful than others? In line with what has been said so far,
the following answer suggests itself. Interpretational success – what brings a
‘candidate’ or potential semantic value into the actual interpretation of the ut-
terance – to a large extent depends on the candidate’s accessibility or degree of
activation. Now a schema is activated by, or accessed through, an expression
whose semantic value corresponds to an aspect of the schema. The schema thus
activated in turn raises the accessibility of whatever possible semantic values
for other constituents of the sentence happen to fit the schema. The schema
itself gains extra activation from the fact that some other constituent of the
sentence has a possible interpretation which fits the schema. In such a case
all ‘candidates’ or potential semantic values which fit the schema evoked by
some of them mutually reinforce their accessibility and therefore increase the
likelihood that they will be globally selected as part of the interpretation of
the utterance. Coherent, schema-instantiating interpretations therefore tend to
be selected and preferred over non-integrated or ‘loose’ interpretations. As a
result, schemata drive the interpretation process.
The role played by schemata explains why the process of utterance interpre-
tation is to such a large extent top down and driven by world knowledge. The
interpreter unconsciously enriches the situation described by the utterance with
many details which do not correspond to any aspect of the uttered sentence but
are contributed in order to fit an evoked schema.21 Thus, in the policeman ex-
ample above, not only is the reference of ‘he’ determined by the s t e a l ( x ) –
i s a r r e s t e d ( x ) schema, but (among other things) the causal interpretation
of the relation between the two events mentioned – the fact that the referent
was arrested because he had stolen – is also determined by the schema.
1 On relevance theory, see Sperber and Wilson, Relevance, Diane Blakemore, Understanding Utter-
ances: An Introduction to Pragmatics (Blackwell, 1992), and Carston, Thoughts and Utterances.
2 See my paper ‘The Alleged Priority of Literal Interpretation’, and more recently ‘Does Linguistic
Communication Rest on Inference?’, in Mind and Language 17 (2002), 105–26.
3 Ruth Millikan, Language, Thought, and Other Biological Categories (MIT Press, 1984),
pp. 60–70; Tyler Burge, ‘Content Preservation’, in Philosophical Review 102 (1993), 457–88.
4 Burge, ‘Content Preservation’, p. 477.
38
Relevance-theoretic objections 39
says. They are arrived at by answering questions such as, ‘Why is the speaker
saying what he says?’ Instead of merely retrieving what is said through the
operation of unconscious, primary pragmatic processes, we reflect on the fact
that the speaker says what he says and use that fact, together with background
knowledge, to infer what the speaker means without saying it. The retrieving
of conversational implicatures therefore involves reflective capacities that are
not exercised in what Millikan calls ‘normal language flow’.5
As I emphasized in chapter 2, the contrast between primary and secondary
pragmatic processes corresponds to different levels of processing. The determi-
nation of what is said takes place at a sub-personal level, like the determination
of what we see. But the determination of what the speaker implies takes place
at the personal level, like the determination of the consequences of what we
see.
Relevance theorists do not accept my distinction between the two sorts of
process. They advocate a unified view, according to which ‘the various different
pragmatic tasks are performed by processes that comprise a single system,
which takes decoded linguistic meaning as its input and delivers the propositions
communicated’.6 And they insist that the processes in question are uniformly
inferential, though in a sense that does not entail reflectiveness.
According to relevance theorists, there is nothing ‘special’ about implica-
tures. They accept that in some cases a conscious, explicit process of inference
takes place, which disrupts normal language flow;7 but it would be a mistake,
they argue, to think of implicature derivation on this model. Implicature deriva-
tion is as automatic and unconscious as the other processes I treat as ‘primary’,
viz. enrichment or loosening. In other words, there are two forms of inference: a
spontaneous, automatic, unconscious form of inference which is involved both
in explicature8 and implicature derivation, and an effortful, conscious form
which is involved only in cases in which something goes wrong and there is
disruption of normal language flow. As Carston puts it,
The appropriate distinction within modes of processing and levels of explanation would
seem to be between, on the one hand, a modular (sub-personal) pragmatic proces-
sor which, when all goes well, quickly and automatically delivers speaker meaning
(explicatures and implicatures), and, on the other hand, processes of a conscious reflec-
tive (personal-level) sort which occur only when the results of the former system are
found wanting.9
inference
conscious unconscious
explicit
effortful
Figure 3.1 A taxonomy of inferences according to Relevance Theory
12 David Marr, Vision: A Computational Investigation into the Human Representation and Pro-
cessing of Visual Information (Freeman, 1968), p. 68.
13 Representations of intensity changes in terms of zero-crossings are clearly non-conceptual (as
pointed out by Tim Crane in The Contents of Experience (Cambridge University Press, 1990)),
yet Marr describes the transition from such proximal representations to distal representations of
edges as an inference. Fodor uses ‘inference’ in the same broad way in The Modularity of Mind
(MIT Press, 1983).
14 ‘[Whenever] the input and the output of a [computation] are not related as premiss and conclusion
in an argument, the computation is not interpretable as inferential’ (Dan Sperber and Deirdre
Wilson, ‘Précis of Relevance: Communication and Cognition’, in Behavioral and Brain Sciences
10 (1987), p. 737).
15 This is debatable. In ‘Does Linguistic Communication Rest On Inference?’, pp. 122–3, I have
questioned the claim that primary pragmatic processes can be described as truth-preserving
operations on conceptual representations.
42 Literal Meaning
16 Sperber, Dan, ‘Intuitive and Reflective Beliefs’, in Mind and Language 12 (1997), pp. 77–8.
17 Ibid., p. 78.
Relevance-theoretic objections 43
inference
conscious unconscious
(sub-personal)
explicit spontaneous
reasoning inference
the first, perceptual belief and the second, inferential belief. Still, the inference,
though not explicit, is conscious in the sense that the availability condition
is satisfied.18 This is in contrast to cases in which the availability condition
is not satisfied: cases, for example, where the subject is aware only of one
judgment, the alleged inferential source of that judgment being unavailable to
consciousness; or cases in which both judgments are available, but the subject
is unaware of one being inferentially derived from the other.
I conclude that there are two sorts of conscious inference: explicit reasoning,
and spontaneous inference (figure 3.2). In both cases the availability condition
is satisfied. Unconscious, sub-personal inferences are characterized by the fact
that the availability condition is not satisfied. Now it is my contention that,
if primary pragmatic processes are indeed ‘inferential’ (in the liberal sense),
the inferences at issue are unconscious and sub-personal. The interpreter is
not aware that his judgment, to the effect that the speaker has said that p, is
inferentially derived from a prior judgment (for example, a judgment to the
effect that the speaker has uttered sentence S). Similarly, some theorists claim
that perceptual judgments themselves are inferential, but the inferences at issue
do not satisfy the availability condition: the perceiver is not aware that his
perceptual judgment that p is based on a prior judgment (to the effect that
18 The distinction between the two notions is crucial. Overlooking it leads to fallacies, as in this
passage from Michael Thau’s Consciousness and Cognition (p. 149): ‘Once we abandon the
idea that conversational participants consciously calculate what’s implicated from what is said,
there’s no reason to think that they should always have the ability to consciously distinguish the
former from the latter.’ This is fallacious because the main reason for denying that conversational
participants consciously calculate what’s implicated from what is said is that they don’t do so
explicitly. (As Thau rightly points out, p. 145, ‘when I understand that your utterance [Fergie
was sober today] is intended to convey that Fergie is often drunk, I don’t have any explicit
thoughts about conversational maxims, nor do I perform any explicit inferential reasoning.’)
But this has no bearing at all on the issue, whether or not the availability condition is satisfied. In
this particular case I take it as obvious that the availability condition is satisfied: the interpreter
is aware that the speaker has said that Fergie is sober today (thereby implying that he is often
drunk).
44 Literal Meaning
things seem thus and so, or whatever).19 In contrast, when I judge that John is
still here because I can see his car, that judgment is transparently or consciously
inferential: the availability condition is satisfied. Similarly, secondary pragmatic
processes are transparently or consciously inferential, because the availability
condition is satisfied: the subject is aware of the relation between the implicature
and what is said. That is true even though (normally) conversational implicatures
are not generated through a process of explicit reasoning. As relevance theorists
rightly emphasize, implicature generation is often as automatic, as effortless as
primary pragmatic processes. But this is irrelevant to the distinction I want to
make. Despite their spontaneous and automatic character, secondary pragmatic
processes are conscious in the minimal sense that the availability condition
is satisfied. In primary pragmatic processes, the availability condition is not
satisfied.
19 Searle gives the following example: ‘Suppose I am standing in front of a house looking at it; in
so doing I will have a certain visual experience with a certain Intentional content, that is certain
conditions of satisfaction; but suppose now as part of the background assumptions I assume I am
on a Hollywood movie set and all of the buildings are just papier maché façades. This assumption
would not only give us different conditions of satisfaction; it would even alter the way the façade
of the house looks to us’ (‘The Background of Meaning’, in John Searle, Ferenc Kiefer and
Manfred Bierwisch (eds.), Speech Act Theory and Pragmatics (Reidel, 1980), p. 231). This sort
of example shows that what we see (in the cognitive sense, involving identification of the seen
object) depends upon background assumptions and can be modified by manipulating them even
though the optical stimuli remain unchanged. This can be described in terms of an ‘inference’
using the relevant assumption as ‘premiss’, but, clearly, the inference at issue is not available to
the perceiving subject.
20 Carston, ‘Linguistic Meaning, Communicated Meaning, and Cognitive Pragmatics’, p. 143.
Relevance-theoretic objections 45
I have already said in what sense I mean that implicatures are special: their
distinctness from what is said is grasped, while the aspects of meaning that are
contextually provided through primary pragmatic processes are phenomenolog-
ically undiscriminated. Be that as it may, I deny that a conversational implicature
is involved in the Gricean example mentioned by Carston. The content of B’s
utterance in (3) is enriched into the proposition which (4) explicitly articulates.
This is very standard. In the relevance-theoretic framework favoured by Carston
herself the word ‘garage’ can be taken to contextually express the ad hoc con-
cept g a r a g e *, paraphrasable as ‘a garage that is presumably open and has
petrol to sell’.24
The inappropriateness of the alleged counterexamples should not blind us to
the important issue raised by relevance theorists: if there are cases in which the
derivation of an implicature must come before the computation of what is said,
does this not argue against my account? To this issue I now turn.
24 Grice’s ‘garage’ example is often treated as the paradigm case of conversational implicature.
In ‘Gricean Rational Reconstructions and the Semantics/Pragmatics Distinction’ (Synthèse 128
(2001), 93–131), Manuel Garcia-Carpintero invokes that example to discredit the ‘Scope Prin-
ciple’: that genuine conversational implicatures (in contrast to pragmatic constituents of what is
said) are not ‘inherited’ when the sentence conveying the implicature is placed under the scope
of logical operators (Direct Reference, pp. 269–74). As against the principle, Carpintero argues
as follows: ‘Uncontroversial cases of implicated contents are also inherited. In another famous
example by Grice, an utterance of ‘there is a gas station around the corner’ implicates that there
is gas available around the corner. This implicated content is inherited in some longer utterances:
just consider ‘if there is a gas station around the corner, I do not need to worry any more’; and
‘there is no gas station nearby’ can be regarded as not falsified by the existence of a closed gas
station around the corner.’ (‘Gricean Rational Reconstructions’, p. 113).
But who said that that example was an uncontroversial case of implicated content? The very
fact that the alleged implicature is inherited in complex sentences makes it controversial that it is
a genuine implicature. (I am indebted to Robyn Carston for mentioning Carpintero’s use of that
example, and this possible line of reply, in her talk ‘Semantics and Conversational Implicature’,
forthcoming in the Proceedings of the Genoa conference WOC 2002: The Semantics/Pragmatics
Distinction, to be published by CSLI Publications.)
25 Tomoko Matsui, ‘Experimental Pragmatics: Towards Testing Relevance-Based Predictions about
Anaphoric Bridging Inferences’, in Varol Akman, Paolo Bouquet, Richmond Thomason and
Roger Young (eds.), Modeling and Using Context (Springer, 2001), p. 250.
Relevance-theoretic objections 47
The implicature here is the premiss that A is not nice. (B’s utterance provides a
relevant answer to A’s question only if this is assumed.) The alleged ‘bridging
implicature’ invoked by Carston as counterexample to my view is also an im-
plicated premiss. That there was beer among the beverages served at the picnic
is an implicated premiss by appeal to which one can enrich ‘the beer’ into ‘the
beer served at the picnic’.
In general I don’t think so-called ‘implicated premisses’ are genuine impli-
catures in the sense in which I am using that term. They are not part of what the
speaker means, that is, intentionally communicates, but part of what he takes for
granted (presupposes) and expects the hearer to take for granted. Thus I draw a
distinction between the common ground to which such premisses belong, and
the explicit or implicit content of the utterance – what the speaker specifically
intends to communicate. Implicatures are, by definition, an aspect of content.
Since that is not true of implicated premisses, they do not count as genuine
implicatures (in my framework).
To be sure, there are cases in which an implicated premiss is part of what is
meant. Thus in the above exchange we may imagine that B’s point precisely
is to convey to A that he (A) is not nice. Even in that case I would not treat
that as an instance of implicature, but, rather, as an instance of ‘staging’ (§5.5).
The speaker ostensively speaks as if it were common ground that A is not
a nice person. The mechanism at work here is more similar to that which
underlies irony than to the inferential mechanism characteristic of implicature.
However, let us assume that there is a genuine implicature here. What, exactly,
is implicated? Wilson says it is the premiss that A is not nice. But there is an
alternative construal. We may consider that what the speaker implicates is that
he considers A as not a nice person. Is this a premiss or a conclusion? Arguably,
like other implicatures it is the conclusion of an inference taking as premiss the
speaker’s saying what he says. The speaker’s saying that only nice persons were
invited, in response to the question ‘Why didn’t you invite me?’, implies that
Relevance-theoretic objections 49
he considers the hearer as not a nice person. If this is an implicature, one may
argue that it is, indeed, grounded in or warranted by the speaker’s locutionary
act (that is, his saying of what is said), just as ordinary implicatures are.
29 ‘In order to derive the implicature, we need the premiss that the speaker has said what he
has said; therefore we need to have identified the primary content of his utterance’ (p. 70);
‘. . . derived from some antecedently determined meaning by an inferential process akin to that
which is involved in conversational implicature’ (p. 72); ‘it is a two-step procedure instead of
being a one-step procedure. The interpreter first determines the utterance’s primary meaning,
then infers some additional meaning’ (p. 74).
30 ‘Report on typescript by François Recanati called Literal Meaning’ (November 2002), p. 3.
31 Garcia-Carpintero, ‘Gricean Rational Reconstructions’, p. 120.
32 Ibid., p. 122. That is the notion of tacit inference appealed to by Scott Soames in Beyond
Rigidity: The Unfinished Semantic Agenda of ‘Naming and Necessity’ (Oxford University Press,
2002). Soames writes: ‘In talking of inference here, I don’t mean to suggest any conscious or
unconscious process of inferring conclusions from premises one consciously or unconsciously
50 Literal Meaning
represents to oneself and takes to be true. What happens is that a sentence is uttered in a context
in which many things are assumed. As a causal result of this, the hearer typically comes to
accept . . . a sequence of sentences . . . The psychological process by which this happens is not
something that can be determined by armchair philosophizing. However, we can construct an
idealized model whose inputs and outputs correspond to those of the real psychological process.
In the idealized model, rational speaker-hearers extract information from utterances by explicitly
representing premises and rationally inferring conclusions from them. When we show that the
inputs and outputs of the real psychological process match those of the idealized model, we
show that our actual methods for gathering information from utterances make sense, and are
rationally justified’ (Beyond Rigidity, p. 335).
33 Here I depart from Carpintero, who thinks the reflective capacity is constitutive of our commu-
nicative competence in general. I will return to that issue in the concluding chapter (§10.3).
34 That does not show that implicatures are not ‘available’, contrary to what Jonathan Berg asserts
(‘In Defense of Direct Belief: Substitutivity, Availability, and Iterability’, in Lingua e Stile 33
(1998), pp. 462–3). Berg uses a simple distinction between conscious and tacit (sub-personal)
inferences and misses the intermediate category unearthed by Carpintero.
4 The Syncretic View
51
52 Literal Meaning
sentence meaning
saturation
what is saidmin
what is saidprag
what is communicated
Figure 4.1 The four-level picture
sense. Salmon, in effect, draws a threefold distinction between (i) what is said
in the minimalist sense (the semantic content of the sentence, in the speaker’s
context), (ii) what the speaker asserts, and (iii) what he or she implies. When
the conventional meaning of the sentence-type is added, this yields something
very much like the four-level picture above.
Kent Bach defends a similar view, with one more level.2 ‘What is said’ is so
minimal, in Bach’s conception, that it need not even be propositional. It may
be a ‘propositional radical’. Thus if, pointing to Tipper, I say ‘He is ready’
what I say is that Tipper is ready. This becomes a full-fledged proposition only
if an answer is provided to the question: ready for what? A pragmatic pro-
cess of ‘completion’ must therefore take place to make the Bachian ‘what is
said’ into a complete proposition. To go from that proposition (resulting from
completion) to what is actually asserted, a further pragmatic process of expan-
sion often comes into play. That process is clearly non-minimalist: it is neither
triggered by a linguistic constituent, nor necessary in order to achieve propo-
sitionality. (‘Expansion’ is Bach’s word for free enrichment.) So in Bach’s
framework there are five distinct notions: the sentence’s linguistic meaning;
what is said in the less-than-minimal sense; the minimal proposition resulting
from completion (= what is said in the minimalist sense); the non-minimal
proposition resulting from expansion (= what is said in the pragmatic sense,
that is, the ‘explicature’, in relevance-theoretic terminology); and, finally, the
Like Salmon and Bach, Soames contrasts what is said in the semantic sense
with what is said in the pragmatic sense (that is, what is stated or asserted). It is
what is said in the pragmatic sense which arguably serves as input to secondary
pragmatic processes and satisfies the availability condition. As we have seen,
that entails that what is said (in the pragmatic sense) cannot satisfy Minimalism.
But nothing follows regarding what is said in the semantic sense; hence there
is no reason to give up Minimalism when dealing with semantic content.
Whatever we think of the details of the views held by Bach, Salmon or
Soames, it is clear that, by freeing us of the limitations of the traditional three-
level picture, they make a reconciliation of Minimalism and Availability possi-
ble. One can be a minimalist with respect to what is said in the strict and literal
sense, and at the same time give up Minimalism when it comes to what the
speaker asserts, which is subject to the availability constraint.
The Syncretic View promises to give us the best of both worlds. It sounds
appealing and reasonable. As Jonathan Berg pointed out, I myself have proposed
something along similar lines.4 Yet I think there is a problem with that view –
a problem which makes it rather unattractive, despite its initial appeal.
The weak point in the Syncretic View is the very notion of ‘what is said in the
strict and literal sense’, that is, the minimalist notion of what is said. How are
we to understand that notion? A natural and widespread interpretation runs as
follows: ‘What is said’ in the minimalist sense is what the sentence says (with
respect to the context at hand), as opposed to what the speaker says by uttering
the sentence. Thus interpreted, however, the Syncretic View is closely related
to a highly dubious, literalist construal of the semantics/pragmatics distinction.
In the next two sections of this chapter, I will spell out that conception and show
what is wrong with it.
3 Soames, Beyond Rigidity, p. 86. 4 Berg, ‘In Defense of Direct Belief’, pp. 466–7.
54 Literal Meaning
occurrence, just as the semantic value of indexicals does, yet it varies not as
a function of some objective feature of the narrow context but as a function
of what the speaker means. It follows that semantic interpretation by itself
cannot determine what is said by a sentence containing such an expression: for
the semantic value of the expression – its own contribution to what is said –
is a matter of speaker’s meaning, and can only be determined by pragmatic
interpretation.
Even if we restrict our attention to expressions traditionally classified as in-
dexicals, we see that they involve a good deal of semantic underdeterminacy.
This is true, in particular, of demonstratives. The reference of a demonstrative
cannot be determined by a rule, like the rule that ‘I’ refers to the speaker. It is
generally assumed that there is such a rule, namely the rule that the demonstra-
tive refers to the object which happens to be demonstrated or which happens to
be the most salient, in the context at hand. But the notions of ‘demonstration’
and ‘salience’ are pragmatic notions in disguise. They cannot be cashed out in
terms merely of the narrow context. Ultimately, a demonstrative refers to what
the speaker who uses it refers to by using it.
To be sure, one can make that into a semantic rule. One can say that the
character of a demonstrative is the rule that it refers to what the speaker intends
to refer to. As a result, one will add to the narrow context a sequence of ‘speaker’s
intended referents’, in such a way that the nth demonstrative in the sentence will
refer to the nth member of the sequence. Formally that is fine, but philosophically
it is clear that one is cheating. We pretend that we can manage with a limited,
narrow notion of context of the sort we need for handling indexicals, while in
fact we can only determine the speaker’s intended referent (hence the semantic
referent, which depends upon the speaker’s intended referent) by resorting to
pragmatic interpretation and relying on the wide context.9
We encounter the same sort of problem even with expressions like ‘here’ and
‘now’ which are traditionally considered as pure indexicals (rather than demon-
stratives). Their semantic value is the time or place of the context respectively.
But what counts as the time and place of the context? How inclusive must the
time or place in question be? It depends on what the speaker means, hence,
9 One may acknowledge the need to appeal to the wide context in interpreting demonstratives
while arguing that the appeal thus made is ‘pre-semantic’. The speaker’s directing intention is
necessary to endow a demonstrative with a character in the first place, one might say. We need the
wide context to fix the expression’s character, but once the character is thus determined, through
a pragmatic procedure akin to disambiguation, it maps the narrow context onto the proposition
expressed by the utterance in that context. (See David Braun, ‘Demonstratives and their Linguistic
Meaning’, in Noûs 30 (1996), 145–73.) Similarly, we might say that whenever there is semantic
underdeterminacy, some form of pragmatic disambiguation must take place before the process
of semantic interpretation can start. To argue in this way, it seems to me, is to acknowledge
that semantic interpretation by itself cannot determine the content of a sentence containing a
semantically underdeterminate expression.
58 Literal Meaning
again, on the wide context. As is well known, ‘here’ can refer to this room, this
building, this city, this country, and so on, and the same underdetermination
affects ‘now’. We can maintain that the character of ‘here’ and ‘now’ is the
rule that the expression refers to ‘the’ time or ‘the’ place of the context – a rule
which automatically determines a content, given a (narrow) context in which
the time and place parameters are given specific values; but then we have to let
a pragmatic process take place to fix the values in question, that is, to determine
which narrow context, among indefinitely many candidates compatible with
the facts of the utterance, serves as argument to the character function. On the
resulting view the (narrow) context with respect to which an utterance is inter-
preted is not given, it is not determined automatically by objective facts like
where and when the utterance takes place, but it is determined by the speaker’s
intention and the wide context. Again, we reach the conclusion that pragmatic
interpretation has a role to play in determining the content of the utterance, in
such a case.
To sum up, either semantic interpretation delivers something gappy, and prag-
matic interpretation must fill the gaps until we reach a complete proposition. Or
we run semantic interpretation only after we have used pragmatic interpreta-
tion to pre-determine the values of semantically underdeterminate expressions,
which values we artificially feed into the narrow context. Either way, semantic
interpretation by itself is powerless to determine what is said, when the sentence
contains a semantically underdeterminate expression.
Now I take it that such expressions can be found all over the place. Moreover,
semantic underdeterminacy is not limited to particular lexical items. One can
follow Waismann and argue that the satisfaction conditions of any empirical
predicate are semantically underdeterminate and subject to pragmatic inter-
pretation (§9.4). There is also constructional underdeterminacy. For example
consider something as simple as the Adjective + Noun construction, as in ‘red
pen’. What counts as a red pen? A pen that is red. But when does a pen count
as red? That depends upon the wide context. The satisfaction conditions of ‘red
pen’ can only be determined by pragmatic interpretation (§9.3).
Suppose I am right and most sentences, perhaps all, are semantically under-
determinate. What follows? That there is no such thing as ‘what the sentence
says’ (in the standard sense in which that phrase is generally used). There is a
single notion of what is said, and that is a pragmatic notion: saying, as Grice
claimed, is a variety of non-natural meaning, characterized by the role which the
conventional meaning of the sentence plays in the hearer’s intended recognition
of the speaker’s communicative intention.
(the literalist interpretation) must be rejected. There is, I claimed, no such thing
as ‘what the sentence says’ in the literalist sense, that is, no such thing as a
complete proposition autonomously determined by the rules of the language
with respect to the context but independent of the speaker’s meaning. As Bach
points out, what the sentence says, in the purely semantic sense, ‘excludes
anything that is determined by [the speaker’s] communicative intention (if it
included that, then what is said would be partly a pragmatic matter)’.10 It is
for that reason that I say there is no such thing: in order to reach a complete
proposition through saturation, we must appeal to the speaker’s meaning. That
is the lesson of semantic underdeterminacy.
There is another possible interpretation for the semantic notion of ‘what is
said’, however. Instead of construing what is said as a non-pragmatic property
of the sentence, independent of the speaker’s meaning, we can start with the
pragmatic notion of what is said, and define the semantic notion in terms of it.
What is said in the minimal sense can thus be defined as what is said in the
full-fledged, pragmatic sense minus the contextual ingredients that are optional
and whose provision is context-driven. To filter out the optional ingredients,
while retaining the contextual ingredients that are necessary for propositionality
(reference of indexicals, and so on), one may follow Soames and define the
semantic content of a sentence s relative to a context C as
that which would be asserted and conveyed by an assertive utterance of s in any normal
context in which the reference of all indexicals in s is the same as their reference in C.11
The strategy now consists in abstracting what is said in the semantic sense from
what is said in the pragmatic sense. That seems possible if, like Soames, we
take the semantic content of the sentence to be included in the content of the
assertion. To get to the semantic content, one only has to filter out those aspects
of assertion content that go beyond semantic content and are tied to specific
contexts of utterance. What remains, that is, the ‘common denominator’, is the
minimal proposition expressed by the sentence itself:
The semantic content of a sentence relative to a context is information that a competent
speaker/hearer can confidently take to be asserted and conveyed by an utterance of the
sentence in the context, no matter what else may be asserted, conveyed, or imparted. It is
a sort of minimal common denominator determined by the linguistic knowledge shared
by all competent speakers, together with contextually relevant facts such as the time,
place, and agent of the context; the identity of individuals demonstrated by the speaker;
and the referents of the names, as used in the context. As such, the semantic content of a
sentence functions as a sort of minimal core around which speaker/hearers can structure
the totality of information the sentence is used to communicate in a given context.12
This alternative strategy also raises a problem, however. What is said in the
minimal, semantic sense cannot be abstracted from what is said in the pragmatic
sense simply because it need not be part of it.13 Soames’s claim that semantic
content is included in assertion content passes muster only because he focuses
on cases in which the asserted content is richer than the alleged semantic content.
He gives the following sort of example:
A man goes into a coffee shop and sits at the counter. The waitress asks him what he
wants. He says, ‘I would like coffee, please.’ The sentence uttered is unspecific in several
respects – its semantic content does not indicate whether the coffee is to be in the form
of beans, grounds, or liquid, nor does it indicate whether the amount in question is a
drop, a cup, a gallon, a sack, or a barrel. Nevertheless, it is obvious from the situation
what the man has in mind, and the waitress is in no doubt about what to do. She brings
him a cup of freshly brewed coffee. If asked to describe the transaction, she might well
say, ‘He ordered a cup of coffee’ or ‘He said he wanted a cup of coffee’, meaning, of
course, the brewed, drinkable kind. In so doing, she would, quite correctly, be reporting
the content of the man’s order, or assertion, as going beyond the semantic content of the
sentence he uttered.14
Soames thinks that in such cases several propositions are asserted, including
(i) the unspecific proposition literally expressed (to the effect that the man wants
coffee in some form or other) and (ii) more specific propositions recoverable
from the literal proposition and the context. Those more specific propositions
resulting from enrichment are tied to the particular context in which they are
generated, hence they can be filtered out by considering other contexts in which
that sentence would be uttered and the indexicals would be given the same
semantic values. Soames equates the minimal proposition expressed by the
sentence with the proposition which would be asserted in all such contexts.
The problem is that enrichment is only one optional process among others.
Only in the case of enrichment, however, is it plausible to suggest that the
minimal proposition itself is part of what is asserted. Soames glosses assertion
in terms of commitment: ‘assertively uttering a sentence with the intention to
assert or convey p involves doing so with the intention of committing oneself
to p’.15 Since one cannot commit oneself to the truth of a specific proposition
p without committing oneself to the truth of a less specific proposition q which
it entails, it makes sense to say that the minimal proposition q is asserted when
one (intuitively) asserts an enriched proposition p. But the principle that the
minimal proposition is part of what is asserted (hence can be abstracted from
it) does not hold when the primary processes at issue are instances of loosening
or transfer. The speaker who assertively utters ‘The ham sandwich left without
paying’, thereby referring to the ham sandwich orderer, does not assert the
minimal proposition that the sandwich itself left without paying! Hence the
minimal proposition cannot be defined as the common denominator – what is
asserted in all contexts in which the sentence is uttered and the indexicals are
given the same semantic values as in the current context.
To be sure, the counterexamples involving loosening or transfer are taken
care of, in Soames’s framework, by the notion of a ‘normal’ context, that is, of
a context in which the sentence ‘is used with its literal meaning’, ‘is not used
metaphorically, ironically, or sarcastically’, and so on. However this qualifica-
tion can hardly be invoked in support of the ‘abstractive’ strategy I am presently
discussing. We are supposed to start with the intuitive (non-minimalist) notion
of what is said, which sometimes is affected by loosening or transfer. If, con-
forming to the abstractive strategy, we want to build the notion of the minimal
proposition out of what is said in that sense, we cannot arbitrarily set aside the
cases that potentially threaten the enterprise.
t h e l i o n ( t h e a n i m a l ) u s e d d u r i n g t h e f i g h t . The minimal
proposition thus determined is absurd and evidently corresponds to no stage in
the actual process of understanding the utterance.
The same sort of argument seems to me to rule out the minimal proposition as
unnecessary. What must ultimately be accounted for is what speakers say in the
pragmatic sense. The job of characters, contents and so on is to contribute to the
overall explanation. But it is sufficient to assign semantic contents (in context)
to simple expressions. Pragmatic processes will operate on those contents, and
the composition rules will compose the resulting pragmatic values, thereby
yielding the content of the speaker’s assertion. Of course it is possible to let the
composition rules compose the semantic contents of the constituent expressions,
16 Jeffrey King and Jason Stanley, ‘Semantics, Pragmatics, and the Role of Semantic Content’,
forthcoming.
17 Ibid.
The Syncretic View 65
thereby yielding the minimal proposition expressed by the sentence. (An absurd
proposition, in the lion’s sword case and many others.) However, the content
of the speaker’s assertion will still be determined by composing the pragmatic
values resulting from the operation of pragmatic processes on the contents of
the constituent expressions; so it is unclear what additional job the minimal
proposition is supposed to be doing.
18 See Robert Stalnaker, Context and Content (Oxford University Press, 1999); John Perry, Ref-
erence and Reflexivity (CSLI Publications 2001), and The Problem of the Essential Indexical,
chapters 11 and 13, and postscript to chapter 2. As Perry points out in The Problem of the Essen-
tial Indexical, p. 246, the notion goes back to Reichenbach’s Elements of Symbolic Logic.
(See also François Recanati, La Transparence et l’Enonciation (Editions du Seuil, 1979),
pp. 161–3.)
66 Literal Meaning
such thing). But this will not do. Our notion of ‘what a sentence says’ is tied to what
speakers who assertively utter the sentence say. Typically, when I assertively utter I am
F, I don’t assert that I am speaking or using language at all. Further, the proposition that
I assert when I assertively utter such a sentence may be true in a possible circumstance
in which no one is using language, and someone may believe this proposition without
believing anything about me being a speaker.19
Soames’s main objection is that the alleged reflexive proposition is not (part
of) what the speaker asserts. As we have seen, however, the same thing often
holds for the minimal proposition posited by the syncretists. The advantage of
the reflexive proposition over the minimal proposition is that it (the reflexive
proposition) is determined solely by the rules of the language, independent
of the speaker’s meaning, in such a way that there is a path to the reflexive
proposition that does not go through the speaker’s meaning; hence it does not
matter much if that proposition can’t be reached by abstraction from what the
speaker asserts.
The reflexive proposition is admittedly distinct from that which the speaker
asserts – they have different possible worlds truth-conditions, as Soames points
out – but why is this an objection? Are we not supposed to draw a sharp dis-
tinction between the proposition expressed by the sentence and the proposition
asserted by the speaker who utters that sentence? Note that we can, if we wish,
incorporate into the reflexive proposition something to the effect that the lin-
guistic mode of presentation associated with the first person pronoun will not
be part of the proposition asserted, while the reference it contextually deter-
mines will be. Thus we might take the reflexive proposition expressed by an
utterance u of ‘I am French’ to be the proposition that there is an x such that
x utters u and u is true iff x is French. This comes as close as one can get to
capturing, in propositional format, the information provided by the utterance in
virtue solely of the linguistic meaning of the sentence ‘I am French.’20 Such a
reflexive proposition determines that the proposition contextually asserted by ‘I
am French’ will consist of the reference of ‘I’ and the property of being French.
The reflexive proposition is therefore ‘tied to what speakers who assertively
utter the sentence say’, even if it is not (part of) what they say.
I conclude that there may be a way of preserving the notion of ‘what the
sentence says’, in the purely semantic sense, if one wants to; but it does not
support the Syncretic View with its four levels. What characterizes the reflexive
proposition is that, although fully propositional, it does not incorporate those
contextual ingredients whose provision is linguistically mandated; it is much
closer to the linguistic meaning of the sentence – indeed it is directly and im-
mediately determined by the linguistic meaning of the sentence. Appealing to
the reflexive proposition instead of the minimal proposition takes us back to
the non-minimalist picture with its three basic levels (figure 1.2): the linguistic
meaning of the sentence (and the reflexive proposition it directly and immedi-
ately determines); what is said in the pragmatic sense; and what is implied or
otherwise conveyed by the utterance.
5 Non-literal uses
It may be thought that the pragmatic approach to what is said advocated in the
first four chapters of this book blurs the commonsensical distinction between
what is (literally) said and what is non-literally conveyed. What is said, in the
pragmatic sense, already incorporates the derived, non-literal values resulting
from primary pragmatic processes of the optional variety. Is there still room for
a contrast between literal and non-literal speech? Wilson and Sperber think this
distinction belongs to folk-linguistics and must be repudiated.1 In this chapter,
however, I will attempt to show that the ordinary, folk-theoretical notion of
non-literal use can be rescued. It corresponds to a range of phenomena which
constitute a natural class, from a phenomenological point of view.
68
Non-literal uses 69
there is nothing in the sentence-type ‘He is thirsty’ that triggers the contextual
generation of the implicature that Paul should be offered a drink.
Let me define: the meaning conveyed by an utterance is ‘m-literal’ iff it
involves only minimal departures from t-literal meaning. (The case in which no
departure is involved may count as a limiting case of m-literalness. I doubt that
there are such cases, however.) Standard cases of non-literalness, in the ordinary
sense, are cases of m-non-literalness: they involve non-minimal departures from
t-literal meaning. Yet, as we shall see, not all cases of m-non-literalness (that
is, not all cases involving such departures) count as ‘non-literal’ in the ordinary
sense.
Consider a third (and last) example, which I already mentioned several times:
(3) You are not going to die.
In the context imagined by Kent Bach, a boy is crying because of a minor
cut and his mother utters (3). What she means is: ‘You’re not going to die
from that cut.’ But m-literally, Bach points out, the utterance expresses the
proposition that the kid will not die tout court. The extra element contextually
provided (the implicit reference to the cut) is not a component of m-literal
meaning: it is not triggered by anything in the sentence, nor is it necessary for
the utterance to express a complete proposition. On the other hand, that element
is not ‘inferred’ on the basis of m-literal meaning: the hearer does not construct
the m-literal interpretation, realize that the utterance, thus interpreted, would
violate Grice’s maxim of quality by entailing something blatantly false (that the
child is immortal), and infer what the speaker actually means; rather, she directly
understands the mother’s utterance as characterizing the situation created by
the cut.5 Once again, the conveyed meaning turns out to be m-non-literal, but
p-literal nonetheless.
can always be made explicit)6 does not take place all the time: it takes place only
when the speaker conveys something indirectly. But processes like enrichment
or loosening are universal: there is no utterance, however explicit, whose inter-
pretation does not involve adjusting the conventional meanings of words to the
particulars of the situation talked about. Communication would be impossible
if things were not so.
The ‘special’ character of non-literal communication is not the whole story,
however. Even if there were something special about m-non-literalness – if,
say, most uses of words were m-literal – a use of words would still not count
as non-literal in the ordinary sense merely in virtue of the fact that it is m-
non-literal. A use of words counts as non-literal in the ordinary sense only if
there is something special about that use that is, or can be, perceived by the
language users themselves. That is so because non-literalness is a feature of the
interpretation of utterances, and the interpretation of utterances is something
that is bound to be available to the language users who do the interpreting. Now
m-non-literalness per se is not transparent to the language users. The speaker and
hearer need not be aware that in their understanding of the uttered words they
are going beyond what the conventions of the language dictate. It is not part of
their competence to reflect upon the complex cognitive processes through which
the meaning of an utterance gets built up from a number of distinct sources.
On the other hand, we have seen that whenever a meaning is accessed via an
inference from a primary meaning, as in (genuine) conversational implicatures,
the language users themselves are aware of the distinction between the two
layers of meaning as well as of the connection between them. In other words,
p-non-literalness is transparent to the language users; hence it counts as non-
literalness in the ordinary sense. (This transparency is not a contingent property
of p-non-literalness. It is definitive of p-non-literalness that the sort of inference
at issue is conducted at the ‘personal’, rather than ‘sub-personal’, level and is
therefore available to the language users.)
To sum up, for something to count as non-literal in the ordinary sense it
must not only go beyond the conventional significance of the uttered words (m-
non-literalness), but it must be felt as such: the language users must be aware
that the conveyed meaning exceeds the conventional significance of the words.
That condition I dub the ‘transparency condition’. It is satisfied whenever the
conveyed meaning has a secondary character, as in conversational implicatures
and indirect speech acts.
5.5 Varieties of non-literal meaning
Is secondariness a necessary, or merely a sufficient condition of transparency?
If it is necessary, then only p-non-literal instances of m-non-literal meaning
6 See section 3.6.
76 Literal Meaning
(conversational implicatures, indirect speech acts, and so on) will count as non-
literal in the ordinary sense. If it is merely sufficient, then presumably there will
be p-literal instances of m-non-literal meaning which will count as non-literal
in the ordinary sense, because the transparency condition has been satisfied
through something other than indirectness. I see no reason to deny that there
are such instances.
The paradigm case of non-literal meaning is metaphor. Now metaphor, in its
most central varieties, I count as p-literal. To re-use an example from chapter 2,
if I say that the ATM swallowed my credit card, I speak metaphorically; there
can be no real ‘swallowing’ on the part of an ATM, but merely something that
resembles swallowing. Still, an ordinary hearer readily understands what is said
by such an utterance, without going through a two-step procedure involving the
prior computation of the ‘literal’ meaning of the utterance (whatever that may
be) and a secondary inference to the actual meaning. Knowing the linguistic
meaning of ‘swallow’, and knowing what sometimes happens with ATMs, the
hearer unreflectively constructs the sense in which the ATM can be said to
‘swallow’ the card by adjusting the meaning of the word to the situation talked
about. This is not very different from what goes on when the meaning of words
is enriched so as to fit the specific situation of discourse, as in the policeman ex-
ample from §5.3. In both cases, as Langacker persuasively argued, the situation
talked about is ‘categorized’ by means of the words which are used to describe
it. In enrichment the situation talked about is a straightforward instantiation of
the schema encoded in the words, which schema therefore gets ‘elaborated’
through its application to the situation. That defines what Langacker calls full
schematicity: ‘the target is compatible with the sanctioning unit (= the word,
with its conventional meaning), and is therefore judged by a speaker to be
an unproblematic instantiation of the category it defines’.7 In contrast, partial
schematicity occurs when ‘there is some conflict between the specifications of
the sanctioning and target structures, so that the former can be construed as
schematic for the latter only with a certain degree of strain’:8
Because partial schematicity involves conflicting specifications, the sanctioning and tar-
get structures cannot merge into a single, consistent conceptualization; in a categorizing
judgement of the form [[SS —> TS]], the discrepancy between SS and TS keeps them
at least partially distinct. The result is a bipartite conceptualization including what we
recognize as a literal sense (SS) and a figurative sense (TS). On the other hand, nothing
prevents the sanctioning and target structures from merging into a unified conceptu-
alization when there is full consistency between their specifications. In the schematic
relationship [[SS —> TS]], SS is in effect ‘swallowed up’ by TS, since all of the spec-
ifications of the former are implicit in the latter, which simply carries them down to a
greater level of precision.9
The picture that emerges is this. As words are applied, in context, to specific
situations, their meaning is adjusted. Depending on whether the conventional
meaning is fully or only partially schematic for the situation talked about, ad-
justment will take one of two forms: sense elaboration (enrichment), or sense
extension (loosening). In sense elaboration the meaning carried by the words
is made more specific through the interaction with contextual factors. In sense
extension, those dimensions of meaning which stand in conflict to the specifi-
cations of the target are filtered out, but they remain somewhat active and may
generate a feeling of discrepancy between the evoked schema and the sense con-
structed by (partially) applying the schema to the situation at hand. That feeling,
like the conflict which underlies it, comes in degrees. Hence there is a contin-
uum between ordinary cases of sense extension that we don’t even perceive
(the ATM swallowing the credit card) and more dramatic cases of metaphor
whose non-literal character cannot be ignored.10 The more noticeable the con-
flict, the more transparent the departure from t-literal meaning will be to the
language users. Beyond a certain threshold, cases of sense extension will there-
fore count as special and non-literal in the ordinary sense, despite their p-literal
character. They will count as figurative uses of language (fig. 5.1).11
Irony and non-serious uses of language are another type of case in which,
I think, the transparency condition is satisfied through something other than
indirectness. To be sure, irony possesses a secondary character (§5.2): by pre-
tending to say of Paul that he is a fine friend in a situation in which just the
opposite is obviously true, the speaker manages to communicate that Paul is
everything but a fine friend. She shows, by her utterance, how inappropriate it
would be to ascribe to Paul the property of being a fine friend. But let us put
aside what is thus implied by saying something ironically – let us concentrate on
the primary meaning of the ironical utterance. In order to understand the utter-
ance at the primary level one must recognize that the act of asserting that Paul is a
fine friend is staged or simulated rather than actually performed. And that means
that one must discern two ‘layers’ within the primary meaning of the utterance:
the surface speech act which the speaker pretends to perform, and the ironical
act of staging the performance of that speech act.12 This layering, internal to the
primary meaning of the utterance, characterizes ‘staged communicative acts’,
10 On this continuum, see Dan Sperber and Deirdre Wilson, ‘Loose talk’. Note that metonymic
transfers also admit of a continuum of cases: some cases count as ‘above threshold’ in the same
sense in which metaphors do. (I am indebted to Robyn Carston for pointing this out to me.)
11 To make the ATM example figurative, we have only to belabour the metaphor so as to make it
conspicuous, as in ‘The ATM swallowed my credit card, chewed it up and spat it out.’ (Kent Bach
suggested this figurative development of the example.) On the distinction between ‘unconscious
metaphorical use of language, as exhibited mainly by young children but also by adults’ and
‘conscious metaphorical use’ in which ‘the awareness of people . . . is caused by the feeling of
discrepancy’, see Bartsch, ‘Word Meanings’, pp. 31–3.
12 Herb Clark, Using Language (Cambridge University Press, 1996), pp. 353–4.
78 Literal Meaning
m-non-literal
p-literal p-non-literal
(indirect speech acts,
conversational
implicatures...)
a large family which includes not only irony but also ‘sarcasm, teasing, over-
statement, understatement, rhetorical questions, and their relatives’.13 Insofar
as the duality of layers is internal to the (primary) meaning of the utterance
and is recognized by whoever understands it, the transparency condition is
eo ipso satisfied: the utterance is intuitively recognized as special, as exhibit-
ing a duality which is absent from standard (‘serious’) uses of language. That
duality, characteristic of staged communicative acts, is a third type of duality,
distinct both from the ‘two-step procedure’ at work in the recovery of secondary
meanings, and from the ‘bipartite conceptualization’ induced by metaphorical
discrepancy. (Of course, nothing prevents a single utterance from exhibiting the
three types of duality, in various patterns of interaction.)
13 Ibid., p. 369. See also Helga Kotthoff, ‘Irony, Quotation, and Other Forms of Staged Intertex-
tuality’, in Carl Graumann and Werner Kallmeyer (eds.), Perspective and Perspectivation in
Discourse (Benjamins, 2002), 201–29.
Non-literal uses 79
have some instances of utterance meaning that are, on the one hand, elements of what is
saidprag (primary meaning), hence the result of subpersonal (unconscious, unavailable)
processes, and are, on the other hand, transparently non-literal, i.e. phenomenologically
salient to speakers/hearers as figurative, i.e. as derived from a more basic meaning, so
that the derivability of the one from the other is consciously available (hence, arguably,
a matter of personal level processing).14
Though I understand Carston’s reaction, I deny that there is a conflict between
the earlier discussion of availability and the discussion of transparency and fig-
urativeness in this chapter. As I have insisted throughout in this book, and again
in section 5.4, ‘the interpretation of utterances is something that is bound to
be available to the language users who do the interpreting’ (see above, p. 75).
This applies both to the primary meaning of the utterance – what is said – and
to the secondary meaning that, in some cases, can be derived from it. Both I
take to be available to the speakers/users (in a normal situation of communi-
cation). What is not available, according to me, is the sub-personal machin-
ery – the processes through which the primary meaning is computed. All that
I maintain.
The cases that Carston sees as problematic I myself describe as cases in
which there is a felt duality internal to the primary meaning of the utterance.
That is not the same thing as the primary/secondary duality, or external duality,
which characterizes implicatures and indirect speech acts. Normal interpreters
are aware of the (secondary) processes through which the secondary meaning
is derived, but they are not aware of the (primary) processes through which
the primary meaning is derived. That is true even if the primary meaning in
question exhibits an internal duality which is transparent to normal interpreters
(figure 5.2).
In my framework, summarized in figure 5.2, the duality which characterizes
figurative uses is internal to the output of the primary pragmatic processes,
and we are aware of it because we are aware of that output. Thus the primary
interpretation of the ‘figurative’ version of the ATM example, ‘The ATM swal-
lowed my credit card, chewed it up and spat it out’15 is a scene in which the
ATM machine exhibits genuine human-like properties. In the scene the ATM
is both a machine and a human being – hence the feeling of discrepancy.16
The discrepancy should not be construed (merely) as a tension between two of
the ingredients which enter into the determination of primary meaning (as in
ordinary cases of sense extension, like the initial, non-figurative ATM exam-
ple, where the very process of sense extension is meant to resolve the tension):
rather, it is a tension between two aspects or components of the primary inter-
pretation. This is the crucial point: the duality which is said to be transparent
14 Carston, ‘Report’, pp. 4–5. 15 See footnote 11, p. 77.
16 See Gilles Fauconnier and Mark Turner, The Way We Think: Conceptual Blending and the Mind’s
Hidden Complexities (Basic Books, 2002) for more examples of metaphorical ‘blending’.
80 Literal Meaning
linguistic meaning
primary
processes unavailable
primary meaning
(possibly exhibiting internal
duality )
secondary
processes external available
duality
secondary meaning
5.7 Conclusion
The literal/non-literal distinction turns out to cover two quite different things:
! For the semanticist, the literal meaning of an expression is the semantic value which
the conventions of the language assign to that expression (‘with respect to context’,
if the expression is indexical). Whenever the meaning which the expression actually
conveys departs from that literal meaning, it is said to be ‘non-literal’.
! In the ordinary sense of the term, non-literal meaning contrasts with normal meaning.
Non-literal meaning is special, it involves a form of deviance or departure from the
norm; a form of deviance or departure which must be transparent to the language
users.
Both distinctions are legitimate. We need both the distinction between con-
ventional meaning and conveyed meaning, and that between normal meaning
and the special meanings assigned to words when the speaker speaks figura-
tively or non-seriously or conveys something indirectly. But it is all too easy to
confuse the two distinctions, which should be kept separate. Indeed I think it is
because the two distinctions have been confused that the same terms have been
used for both distinctions.
The culprit here is the tacit, but very widespread, assumption that the m-literal
meaning of words is what, in context, they normally express. In semantics (as
opposed to psychology), departures from m-literal meaning tend to be treated
as special, on the pattern of figurative language, non-serious speech or indirect
communication. I think that assumption is mistaken. In context, words system-
atically express meanings that are richer than (or otherwise different from) what
the conventions of the language dictate. Contrary to what formal semanticists
tend to assume, the (intuitive) truth-conditions of our utterances are not compo-
sitionally determined by the meanings of words and their syntactic arrangement,
in a strict bottom-up manner. They are shaped by contextual expectations and
world-knowledge to a very large extent. That is true of all utterances, however
82 Literal Meaning
‘literal’ they are (in the ordinary sense). If we abstract from those top-down
factors, what we get – the utterance’s m-literal meaning – no longer corresponds
to the intuitive truth-conditions which the language users themselves associate
with their own utterances. In other words, there is a gap between the deliver-
ances of semantic theory and the intuitive content of utterances. That gap is
bridged by pragmatic processes which take place normally and do not generate
‘non-literal meanings’, except in special cases.
6 From Literalism to Contextualism
1 See §10.1 below (pp. 154–8), for a discussion of two famous anti-contextualist arguments.
83
84 Literal Meaning
Contextualists2 held that the difference between the two types of language was
all-important; natural-language sentences, according to them, were essentially
context-sensitive, and did not have determinate truth-conditions. Literalists, on
the other hand, believed that the difference between the two types of language
could be abstracted from through a legitimate idealization.
The literalist idealization was based on the following claim:
(L) For every statement that can be made using a context-sensitive sentence in a given
context, there is an eternal sentence that can be used to make the same statement in
any context.
2 The contextualists I am talking about here are the so-called ordinary language philosophers:
Wittgenstein, Austin, Strawson among others.
3 Note that (L) is much weaker than another principle of effability, namely (T):
(T) Every entertainable thought may be expressed by means of an eternal sentence the sense of
which corresponds exactly to that thought.
Many philosophers have (rightly) argued against (T). For example, Sperber and Wilson write: ‘It
seems plausible that in our internal language we often fix time and space references not in terms
of universal coordinates, but in terms of a private logbook and an ego-centred map; furthermore,
most kinds of reference – to people and events for instance – can be fixed in terms of these
private time and space coordinates. Thoughts which contain such private references could not be
encoded in natural languages but could only be incompletely represented’ (Relevance, p. 192).
(L) is not subject to that Perry-type criticism. (L) says only that every statement can be made
using an eternal sentence, not that every thought can be literally expressed by an eternal sentence.
Now, a statement may be of an object, in the sense that it may be about a certain object without
involving a particular mode of presentation of that object. Such a de re statement corresponds to
a class of thoughts, each involving a particular (and, perhaps, private) mode of presentation of the
object referred to. The fact that, in the thought, there are private modes of presentation attached
to the objects referred to implies that there are thoughts that cannot be totally and adequately
represented by means of eternal sentences, but does not imply that there are statements that cannot
be made by means of eternal sentences.
From Literalism to Contextualism 85
Table 6.1
! Literalism "
! Indexicalism literalist side
! The Syncretic View
! Quasi-Contextualism # contextualist
! Contextualism side
acknowledges such influences, but limits their effects to what is said in the
intuitive sense, as opposed to what is strictly and literally said (the minimal
proposition).
The next two positions I want to consider fully acknowledge the role of
pragmatic processes, including pragmatic processes of the optional variety, in
the determination of what is said. What is said is taken to be pragmatically
determined, in a way that is sensitive to top-down influences. In contrast to
the Syncretic View, those positions do not postulate a level of content that is
unaffected by top-down factors. So they fall on the contextualist side. They differ
from each other in their respective attitudes towards the minimal proposition
posited by the Syncretic View. One position, which I call Quasi-Contextualism,
simply considers the minimal proposition as a theoretically useless entity which
plays no role in communication (§4.5). Full-fledged Contextualism goes further
and denies that the notion even makes sense.
To sum up, the five positions in the ballpark are shown in table 6.1. Literalism
and the Syncretic View have been explicitly criticized, and I have implicitly en-
dorsed Quasi-Contextualism in arguing against the Syncretic View in chapter 4.
So two positions remain to be introduced and discussed: Indexicalism, and
Contextualism.
6.2 Indexicalism
The pragmatics literature is replete with examples in which what is said in the
intuitive sense seems to be affected by pragmatic processes of the optional (top-
down) variety – what King and Stanley call ‘strong pragmatic effects’.4 Yet, as
Stephen Levinson (one of the providers of examples) puts it, ‘there will always
be doubt about whether a better semantic analysis of the relevant construction
might not accommodate the apparent pragmatic intrusion in some other way’.5
Indeed, in most cases, there are different ways of analysing a given example.
Some of those analyses may well suggest that the pragmatic process at issue is
not a primary pragmatic process of the optional variety, but a regular instance
of saturation.
4 See King and Stanley, ‘Semantics, Pragmatics, and the Role of Semantic Content’.
5 Stephen Levinson, Presumptive Meaning, p. 214.
From Literalism to Contextualism 87
The suggestion is that, when the sentence is uttered, a value has to be contex-
tually assigned to the variable, thereby restricting the domain of quantification.
If we take this line, we treat quantifier domain restriction as, in effect, a case
of saturation.6 In this framework, the cases in which the quantifier phrase is
given an ‘absolute’ reading are treated as cases in which a value is contextu-
ally assigned to the variable, but intuitively no restriction takes place because
the contextually selected domain is the whole universe (or the entire domain
relevant to the variable, in case we use sorted variables).
Or consider the examples of enrichment with which our discussion of Mini-
malism began:
I’ve had breakfast <this morning>
You’re not going to die <from that cut>
6 See for example, Jason Stanley and Zoltán Szabó, ‘On Quantifier Domain Restriction’, in Mind
and Language 15 (2000), 219–61.
88 Literal Meaning
As I pointed out in ‘The Pragmatics of What is Said’, this analysis faces dif-
ficulties when the past tense is replaced by the present perfect, as in Jonathan
Cohen’s example ‘The old king has died of a heart attack and a republic has
been declared.’13 That is so because the present perfect can hardly be consid-
ered as referring to a specific time. But if, plausibly enough, we construe the
present perfect as quantifying over times, we can extend the analysis in terms
of quantifier domain restriction to that type of example also. The sentence will
be understood as saying that there is a time t in the past such that the old king
dies of a heart attack at t, and there is a time t’ in the past such that a republic
is declared at t’. If the domain to which the second quantifier is indexed is con-
textually restricted to the set of times that are posterior to t, we account for the
relevant temporal reading. So it seems that there always is a way of analysing
Gricean examples of temporally understood conjunctions in terms of saturation
rather than enrichment.
Examples can be multiplied at will. The indexicalist position consists in
systematically offering an analysis in terms of saturation, whenever the intuitive
truth-conditions of an utterance are affected by contextual factors. As Jason
Stanley, the chief representative of this research program, puts it, ‘all effects of
extra-linguistic context on the truth-conditions of an assertion are traceable to
elements in the actual syntactic structure of the sentence uttered’.14 To maintain
such a sweeping claim, one must deal with every putative counterexample, in
order to establish, in each case, the conclusion that ‘we have been given no
reason to abandon the thesis that the only truth-conditional role of context is
the resolution of indexicality, broadly construed’.15
At this point the debate becomes technical (and, fortunately, empirical). It’s
a matter of detailed analyses of particular examples (or classes of examples).
Without going into those details, it is fair to say that the indexicalist starts with a
significant disadvantage; for he makes a universal claim while his opponent only
makes an existential claim. For his opponent to win, it is sufficient to produce
6.3 Contextualism
According to Contextualism – a provocative view which certainly deserves
to be explored – there is no level of meaning which is both (i) propositional
(truth-evaluable) and (ii) minimalist, that is, unaffected by top-down factors.
As I put it in Direct Reference, ‘no proposition could be expressed without
some unarticulated constituent being contextually provided’.16 Such a radical
view has been defended by Charles Travis, John Searle, Julius Moravcsik and
a few others. Searle, for example, argues that a determinate proposition can be
expressed only against a background of unarticulated assumptions. He gives
the following example of unarticulated assumptions at work in understanding
a simple utterance:
Suppose I go into the restaurant and order a meal. Suppose I say, speaking literally,
‘Bring me a steak with fried potatoes.’ (. . .) I take it for granted that they will not deliver
the meal to my house, or to my place of work. I take it for granted that the steak will not
be encased in concrete, or petrified. It will not be stuffed into my pockets or spread over
my head. But none of these assumptions was made explicit in the literal utterance.17
Though unarticulated, those assumptions contribute to determining the intuitive
conditions of satisfaction (obedience-conditions, truth-conditions, and so on) of
the utterance. The order ‘Bring me a steak with fried potatoes’ does not count as
satisfied if the steak is delivered, encased in concrete, to the customer’s house. It
is mutually manifest to both the hearer and the speaker that the speaker intends
the ordered meal to be placed in front of him on the restaurant table he is sitting
at, and so on.
In that sort of case a syncretist will insist that what is strictly and literally
said is free from the relevant assumptions: the order ‘Bring me a steak with
fried potatoes’ would be literally satisfied if the steak was delivered, encased
in concrete, to the customer’s house. (See Soames’s treatment of the ‘coffee’
example in §4.4.) That is precisely what the contextualist denies. Another ex-
ample given by Searle will help to make sense of the contextualist position. The
16 Recanati, Direct Reference, p. 260.
17 John Searle, The Rediscovery of the Mind (MIT Press, 1992), p. 180.
From Literalism to Contextualism 91
word ‘cut’ is not ambiguous, Searle says, yet it makes quite different contribu-
tions to the truth-conditions of the utterance in ‘Bill cut the grass’ and ‘Sally
cut the cake’. That is because background assumptions play a role in fixing
satisfaction-conditions, and different background assumptions underlie the use
of ‘cut’ in connection with grass and cakes respectively. We (defeasibly) as-
sume that grass is cut in a certain way, and cakes in another way. The assumed
way of cutting finds its way into the utterance’s truth-conditions:
Though the occurrence of the word ‘cut’ is literal in (both) utterances . . . and though the
word is not ambiguous, it determines different sets of truth-conditions for the different
sentences. The sort of thing that constitutes cutting the grass is quite different from, e.g.,
the sort of thing that constitutes cutting a cake. One way to see this is to imagine what
constitutes obeying the order to cut something. If someone tells me to cut the grass and
I rush out and stab it with a knife, or if I am ordered to cut the cake and I run over it
with a lawnmower, in each case I will have failed to obey the order. That is not what the
speaker meant by his literal and serious utterance of the sentence.18
Now the syncretist will insist that a sentence such as ‘Cut the grass’ expresses
something that has literal conditions of satisfaction quite independent of any
background assumption; something very abstract, involving the constant, un-
derspecified meaning of ‘cut’ and not the definite sense it takes on particular
uses (or types of use). Stabbing the grass with a knife and running over it with
a lawnmower are two ways of literally obeying the order ‘Cut the grass’, on
this view. But the contextualist remains sceptical. To get something genuinely
evaluable, he claims, that is, something which enables us to partition possible
worlds into those in which the relevant condition is satisfied and those in which
it is not, we need background assumptions. We cannot specify a determinate
proposition which the sentence can be said literally to express, without building
unarticulated assumptions into that proposition. The best we can do is to con-
struct a disjunction of the propositions which could be determinately expressed
by that sentence against alternative background assumptions.
In support of this claim, Searle sets up an example for which no background
assumption is readily available: ‘Cut the sun.’ What counts as obeying that
order? We don’t quite know. The abstract condition we can associate with that
sentence (involving some form of linear separation affecting the integrity of
the sun) is, precisely, too abstract to enable us to tell the worlds in which the
condition is satisfied from the worlds in which it is not. It is not determinate
enough to give us specific truth-conditions or obedience-conditions.19
(i) Gary Cooper (the speaker) is handing a fricassee plate to the waiter (Pedro); (ii) the fricassee
plate contains a gun; (iii) that gun has just fallen from the hands of someone during a brief fight
around the dinner table. With respect to that situation, the phrase ‘disarm the fricassee’ makes
sense: it means that the waiter is to remove the gun from the plate. Without a proper background,
however, we no more know the obedience conditions of Cooper’s utterance ‘Disarm the fricassee’
than we know the obedience conditions of ‘Cut the sun.’
20 As King and Stanley point out in their forthcoming paper ‘Semantics, Pragmatics, and the Role
of Semantic Content’, Syncretism – what they call ‘semantic modesty’ – ‘is a position that
comes in degrees. On the one extreme, there are those who hold that most of our intuitions
about what is said by a sentence in a context are affected by strong pragmatic effects. On the
other end of the spectrum, are theorists who hold that while generally intuitions about what is
said by a sentence are reliable guides to semantic content, there are a restricted range of cases
in which strong pragmatic effects may affect speaker intuitions about what they take to be the
semantic content of a sentence.’ Cappelen and Lepore are radical syncretists. They argue for
systematically disconnecting the semantic content which the theorist assigns to a sentence from
what is intuitively said by uttering that sentence (see Cappelen and Lepore, ‘On an Alleged
Connection Between Indirect Speech and the Theory of Meaning’).
21 Borg, ‘Saying What You Mean’.
22 Ibid. The first occurrence of this line of reply to contextualism can be found in Marcelo
Dascal’s discussion of Searle’s ‘Literal Meaning’: see Marcelo Dascal, ‘Contextualism’, in
Parret, Sbisà and Verschueren (eds.), Possibilities and Limitations of Pragmatics, pp. 173–4.
The most recent occurrence I have seen is in Herman Cappelen and Ernie Lepore, ‘Radical and
Moderate Pragmatics: Does Meaning Determine Truth-Conditions?’ (forthcoming).
From Literalism to Contextualism 93
31 In ‘The Pragmatics of What is Said’, pp. 308–9 n., I argued that the Minimalist Principle can
serve as a methodological weapon in the hand of the contextualist.
From Literalism to Contextualism 97
98
Indexicalism and the Binding Fallacy 99
of the original relation. As Quine pointed out, from any n-place predicate P,
one can generate an n − 1 place predicate by applying to P an operator he calls
‘Derelativization’, which he describes as follows:
(Der P) x1 . . . xn−1 iff there is something xn such that P x1 . . . xn
If ‘P’ is a two-place predicate, ‘Der P’ will be a genuine one-place predicate,
denoting a property rather than a relation.2 In the case of ‘eat’ and other verbs
subject to INI, suppressing the direct object in surface syntax amounts seman-
tically to the same result as applying ‘Der’ to the original two-place predicate.
Intransitive ‘eat’ thus denotes the property of eating (e at 1 ), namely the prop-
erty one has in virtue of filling the e at e r argument-role in some instance of
the dyadic e at relation (e at 2 ). One eats1 iff there is something that one eats2 .
In definite null instantiation (DNI) the situation is quite different. Like ‘eat’,
‘hear’ and ‘notice’ are transitive verbs which have an intransitive use: they can be
used without a direct object, as in ‘I heard’ or ‘I noticed’. When that is the case,
however, there must be something definite, in context, which the speaker is said
to have heard or noticed. To understand what is said, an interpreter must identify
that thing. This shows that the objectless verb still denotes the original two-
place relation, even though the second argument-role is not realized in surface
syntax.3 Unless a particular filler is contextually assigned to the unrealized
argument role, the utterance remains semantically incomplete. Thus ‘I heard’
or ‘I noticed’ cannot be interpreted as meaning that the speaker has heard or
noticed something or other. This is what it means to say that, in contrast to
‘eat’, ‘hear’ is not subject to INI, but to DNI. ‘John eats’ is true iff there is
something that John eats2 . The fact that the relevant thing does not have to be
specified supports the conclusion that intransitive ‘eat’ denotes the property of
eating1 . In contrast, ‘x noticed’ is definitely not true iff there is something y
which x noticed. Rather, it is true, for some contextually specified thing y, iff
x noticed that thing. The relevant y has to be specified for the utterance to be
truth-evaluable. That is an instance of saturation, even though the constituent
to be contextually provided is ‘unarticulated’ in surface syntax.4
2 Willard van Orman Quine, ‘Variables Explained Away’, in his Selected Logic Papers (Harvard
University Press, 1995), pp. 229–31.
3 In the situation-theoretic framework I will sketch in chapter 8, an alternative account of DNI is
available. The missing argument can be construed as belonging to the ‘circumstance of evaluation’
rather than to the ‘articulated content’. Thus, instead of maintaining that the objectless verb still
denotes the original two-place relation, we can say that the objectless DNI-verb is an indexical
one-place predicate: the property it denotes is relativized to, and varies with, the ‘circumstantial’
constituent that is contextually provided. Thus in a context c1 intransitive ‘notice’ will denote
the property of noticing the accident, while in another context c2 it will denote the property of
noticing the man. See my ‘Unarticulated Constituents’, p. 312, footnote 11 for an analogous
suggestion concerning another type of implicit argument.
4 Insofar as it is mandatory and provided through saturation, the constituent is unarticulated only
in a weak sense. For the distinction between two senses of ‘unarticulated’, see my ‘Unarticulated
Constituents’, pp. 307–13.
100 Literal Meaning
To sum up, it is a conventional property of the English verb ‘eat’ that (i) it
can be used intransitively (in contrast to ‘devour’), and (ii) when so used it does
not require saturation of the unarticulated argument-role. That argument role
is best thought of as suppressed: the two-place relation has been replaced by a
property. Likewise, it’s a conventional property of the English verb ‘finish’ that
(i) it too can be used intransitively (in contrast to ‘complete’), but (ii) when so
used it does require saturation of the unarticulated argument-role. The latter is
not suppressed, though it remains implicit.
The fact that the suppressed argument role of intransitive ‘eat’ does not
require saturation does not entail that a value cannot be contextually provided for
that role. What the DNI/INI contrast suggests is only this: if a value is provided
for the suppressed argument role of ‘eat’, it is provided through an optional
pragmatic process of enrichment. This is different from the mandatory provision
of a contextual value for the implicit argument-role in need of saturation, in
cases of DNI such as ‘finish’ or ‘notice’. Verbs subject to DNI carry something
like a free variable in semantic structure, which must be assigned a definite
value in context. Verbs subject to INI do not carry such a free variable, hence
the contextual provision of a value for the suppressed argument role cannot be
treated as an instance of saturation, but as a case of enrichment.
Let me give an example. We can imagine a context in which ‘Look! He’s
eating!’ would be understood as stating not merely that the individual denoted
by ‘he’ is eating something or other, but that he is eating a certain salient
piece of food, for example, a dangerous mushroom which has been the focus of
attention for some time. The eaten object would then be contextually provided
without being linguistically required, since the intransitive verb only denotes
the property and does not require saturation. In such a case the contextual
specification of the eaten object would result from free enrichment. But in
the above cases of DNI (‘I heard’, ‘I finished’, ‘I noticed’), the contextual
provision of the relevant ingredient is a bottom-up pragmatic process – a variety
of saturation – rather than a top-down pragmatic process of free enrichment.
This is established by the fact that the provision of an object filling the second
argument role is required in every context, not just in the contexts in which it
turns out to be relevant.
Optionality Criterion
Whenever a contextual ingredient of content is provided through a pragmatic process of
the optional variety, we can imagine another possible context of utterance in which no
such ingredient is provided yet the utterance expresses a complete proposition.
Systematically used, that criterion may give surprising results. Thus it is com-
mon to hold that the sentence ‘It’s raining’ expresses a complete proposition
only if a location is contextually provided. If that’s right, then the contextual
specification of the location of rain is an instance of saturation, much like the
contextual provision of the thing heard in ‘I heard’.5 Using the Optionality
Criterion, however, I have established that the location of rain is not provided
through saturation; for there are contexts in which the sentence ‘It is raining’
expresses a complete proposition, even though no location is contextually pro-
vided as that which the utterance concerns (see §1.3, pp. 9–10).
Applying the Criterion is not always easy, however. For it is not always easy
to tell whether or not a contextual ingredient is actually provided. In §6.2 I
gave an example of that difficulty. There are two ways to look at contextual
restrictions of the domain of quantification. First possibility: we say that a
contextual ingredient is provided only when the domain of quantification is
implicitly restricted. The cases in which the quantifier ranges over the entire
domain of discourse, without contextual restriction, are treated as cases in which
no pragmatic process takes place – the sentence is simply taken at face value.
The fact that there are such ‘absolute’ readings of quantified sentences can
then be considered as showing that the process of contextual domain restriction
is not mandatory but optional. Second possibility: we consider that a primary
pragmatic process of saturation must take place to assign a value to the domain
variable in logical form. Absolute readings themselves result from a particular
assignment of value to the variable, but that is not visible because of the nature
of the assignment: the entire domain is assigned to the variable, hence no
contextual restriction intuitively occurs. A contextual ingredient is nonetheless
provided. The Optionality Criterion, by itself, cannot help us with the choice
between the two possibilities. It’s only after the choice has been made that we
can apply the Optionality Criterion. That is, we must decide, first, whether or
not a contextual ingredient is provided when the quantified sentence is given
5 Thus Ken Taylor writes (‘Sex, Breakfast, and Descriptus Interruptus’, p. 53): ‘The view which
I favor supposes that the verb ‘to rain’ has a lexically specified argument place which is #–
marked t h e m e and that this argument place takes places as values. This is a way of saying
that the subatomic structure of the verb “to rain” explicitly marks rainings as a kind of change
that places undergo. (. . .) Thus though “It is raining” is missing no syntactically mandatory
sentential constituent, nonetheless, it is semantically incomplete. The semantic incompleteness
is manifest to us as a felt inability to evaluate the truth value of an utterance of [that sentence]
in the absence of a contextually provided location (or range of locations). This felt need for a
contextually provided location has its source, I claim, in our tacit cognition of the syntactically
unexpressed argument place of the verb “to rain”.’
102 Literal Meaning
the ‘absolute’ reading. It’s only if we give a negative answer that we can use
the Optionality Criterion to establish that quantifier domain restriction is not an
instance of saturation.
Are there other criteria? According to Jason Stanley, there is one that is
most helpful: the Binding Criterion. Stanley’s idea is the following. Whenever
a contextual ingredient is provided as a result of saturation, there is something
like a free variable in logical form, which must be contextually assigned a
value. If a contextual ingredient does not result from saturation, but is pragmatic
through and through, there is no variable in logical form to which that ingredient
corresponds. It is ‘unarticulated’ in the strongest possible sense. But that entails
that no ‘binding’ of the relevant constituent is possible:
Since the supposed unarticulated constituent . . . is not the value of anything in the
sentence uttered, there should be no reading of the relevant linguistic constructions in
which the unarticulated constituent varies with the values introduced by operators in the
sentence uttered. Operators in a sentence only interact with variables in the sentence
that lie within their scope. But, if the constituent is unarticulated, it is not the value of
any variable in the sentence. Thus, its interpretation cannot be controlled by operators
in the sentence.6
For binding to occur, Stanley argues, there must be a bindable variable in the
sentence to which the operator is prefixed; but if there is such a variable, then
the contextual ingredient (when there is one) ought to be seen as the value
contextually assigned to that variable in the process of saturation, rather than
as an ‘unarticulated constituent’ corresponding to nothing in the sentence and
resulting from ‘free enrichment’. This, then, is the criterion Stanley uses:
Binding Criterion
A contextual ingredient in the interpretation of a sentence S results from saturation if
it can be ‘bound’, that is, if it can be made to vary with the values introduced by some
operator prefixed to S.
Using that criterion, Stanley attempts to show that quantifier domain restric-
tion is an instance of saturation. For there are cases in which the domain of a
quantifier may be bound by another quantifier, as in the following examples:
(1) In most of John’s classes, he fails exactly three Frenchmen.
(2) In every room in John’s house, every bottle is in the corner.
(3) Whatever John does, most of the class falls asleep.
in x. Thus the domain of the quantifier expression ‘three Frenchmen’ varies with the
value of the variable introduced by the quantifier ‘most’.7
Since there is no binding without a bindable variable, Stanley and Szabó con-
clude that ‘in the logical form of quantified sentences, there are variables whose
values, relative to a context, are (often restricted) quantifier domains’.8 It fol-
lows that the contextual provision of a restricted domain of quantification is an
instance of saturation.
The Optionality Criterion tells us that the contextual provision of the comparison
class (for ‘small’) or of the thing noticed (for intransive uses of ‘notice’) is in
both cases an instance of saturation because there is no possible context of
utterance for (4) or (5) in which no such ingredient would be provided yet
the sentence would still express a complete proposition. Using the Binding
Criterion we reach the same conclusion, for binding can occur in both cases.
Stanley gives the following example:
(6) Most species have members that are small.
Here, instead of being contextually specified, the comparison class varies with
the values introduced by ‘most species’: each species in turn serves as com-
parison class for smallness. Likewise, Partee points out that we can bind the
‘object’ of ‘notice’ even though it is unarticulated in surface syntax:9
(7) Every secretary made a mistake in his final draft. The good secretary corrected his
mistake. Every other secretary did not even notice.
7 Stanley and Szabó, ‘On Quantifier Domain Restriction’, p. 243. 8 Ibid., p. 258.
9 See Barbara Partee, ‘Binding Implicit Variables in Quantified Contexts’, CLS 25 (1989), 342–65.
104 Literal Meaning
and mean that each of the persons in question went to his or her home. On
that interpretation the implicit argument varies with the values introduced by
‘everybody’. According to the Binding Argument, this shows that there is a
variable in logical form, which can be bound or contextually assigned a value,
as the case may be. When a value is contextually assigned, as in (8), this is an
instance of saturation.
Life is not as simple as that, however. There are also cases in which the two
criteria yield conflicting results. By the Optionality Criterion, as we have seen,
the location of rain which is implicit in ‘It’s raining’ is provided through an
optional process of free enrichment; yet, Stanley points out, it can be bound:
(10) Everywhere I go, it rains.
In such a statement the location of rain is undertood as varying with the places
introduced by the quantifier ‘Everywhere I go’. The statement means that:
For every location l such that I go to l, it rains in l (when I am there).
For such binding to occur, Stanley says, there must be a free variable l in
the sentence ‘it rains’. That variable can either be bound (as in (10)), or be
contextually given a value. Whatever location may be contextually provided
for the rain is therefore not a genuine ‘optional’ ingredient, but the contextual
value of a free variable in logical form. The contextual provision of a specific
location therefore counts as an instance of saturation, contrary to the verdict
delivered by the Optionality Criterion.
Similarly, the temporal location of the breakfast event in Sperber and Wilson’s
example (‘I’ve had breakfast’) seems to result from enrichment according to
the Optionality Criterion; yet it can be bound by a quantifier.10 We can say:
(11) No luck. Each time you offer me lunch, I’ve had a very large breakfast.
The temporal location of the breakfast event now systematically varies with the
temporal values introduced by ‘each time you offer me lunch’. It follows that
there is a variable in logical form. But if there is, then the alleged instance of
enrichment is an instance of saturation after all. So the argument goes.
Unsurprisingly, Stanley uses the Binding Criterion in arguing for Indexical-
ism. He appeals to data of the sort originally collected by Barbara Partee11 to
show that, whenever an ‘optional’ ingredient has been postulated to account for
10 Stanley does not actually discuss that example, but his general argument applies nonetheless.
11 Ibid. See also Jonathan Mitchell, The Formal Semantics of Point of View (PhD dissertation,
University of Massachusetts, Amherst, 1986), where some of Partee’s examples originate. An
early discussion of the quantification of implicit variables can be found in Östen Dahl, ‘On
Points of Reference’, in Semantikos 1 (1975), pp. 58–60. Dahl refers the reader to an unpublished
manuscript by George Lakoff, dated 1970.
Indexicalism and the Binding Fallacy 105
the intuitive meaning of an utterance, one can intuitively ‘bind’ it, that is, make
it vary according to the values introduced by some operator. It follows that ‘we
have been given no reason to abandon the thesis that the only truth-conditional
role of context is the resolution of indexicality, broadly construed’.12 Stanley
therefore maintains that ‘all truth-conditional context-dependence results from
fixing the values of contextually sensitive elements in the real structure of nat-
ural language sentences’.13 There are no ‘optional’ contextual ingredients in
truth-conditional content.
If we apply the Binding Criterion, we shall have to conclude that the contextually
provided manner of stopping is articulated and determined through a bottom-
up process of saturation, like the reference of indexicals. The absurdity of this
conclusion argues against the Binding Criterion.
The other example is even more striking. Remember the utterance: ‘Look! He
is eating!’ We imagined a context in which a salient mushroom was understood
as being the thing eaten. That the contextually provided constituent results
from free enrichment follows from the fact that intransitive ‘eat’, as Stanley
himself accepts,15 denotes the property of eating, abstracted from the relation
of eating-something by suppressing the second argument role. That is a case of
‘indefinite null instantiation’ in which no contextual specification of the thing
eaten is required in virtue of the semantics of the verb (in contrast to what
happens in cases of ‘definite null instantiation’, such as ‘I noticed’ or ‘I heard’).
Still, intuitively, binding is possible. We can say:
(15) John is anorexic, but whenever his father cooks mushrooms, he eats.
On a natural interpretation, we understand that John eats the mushrooms his
father has cooked. Intuitively, a form of binding is operative here; for the food
eaten by John covaries with the food cooked by his father. Such examples show
that intuitive binding, per se, does not entail the existence of a free variable in
15 See Stanley, ‘Context and Logical Form’, p. 401, fn. 14.
Indexicalism and the Binding Fallacy 107
logical form. The Binding Criterion, on which Stanley’s argument rests, must
be rejected.
It remains to be shown what exactly is wrong with the Binding Criterion, and
with Stanley’s argument in support of Indexicalism. I will provide a detailed
analysis of what I call the ‘Binding Fallacy’ in §7.6. First, however, we need a
bit of stage-setting, which the next section will provide.
There are also ‘recessive’ variadic functions that decrease the valence of the
input relation by suppressing an argument-role.17 In English, various alterna-
tions such as the passive alternation and the intransitive alternation18 can be
analysed in such terms. The operation of passivation (‘John kisses Mary’ →
‘Mary is kissed’) suppresses the argument-role corresponding to the subject of
the active sentence, whereas intransitivation (‘John eats the apple’ → ‘John
eats’) has the effect of suppressing the argument-role corresponding to the
direct object of the verb. Quine’s ‘Der’ operator, which I mentioned earlier,
does something similar: it decreases the adicity of the input relation Rn by
existentially quantifying the nth argument. But I think recessive variadic func-
tions do not, by themselves, involve existentially quantifying the suppressed
argument-role. Thus if we start with ‘John spoke to Mary’, we can suppress the
argument-role of a d d r e s s e e and generate ‘John spoke’. This does not entail
that John spoke to someone. Whether or not the suppression of an argument-
role, effected through a recessive variadic function, amounts to existentially
quantifying that argument-role depends upon the nature of the argument-roles
that remain unsuppressed. To fill the s p e a k e r argument-role one need not
have an addressee. One can speak without speaking to anyone. But to fill the
e at e r argument-role, arguably, there must be something (edible) that one
eats: n o e at e r without an e at e e . The suppressed argument-role therefore
remains in the background via the e at e r argument-role which, for metaphys-
ical reasons, cannot be entirely dissociated from it. That is why ‘John is eating
something’ can be inferred from ‘John is eating.’ That, according to me, is a
non-formal, ‘metaphysical’ inference similar to that from ‘Mary is dancing’ to
‘Mary is dancing somewhere.’19
17 ‘Recessive’ comes from Lucien Tesnière, a pioneer in that area (Eléments de Syntaxe Struc-
turale, 2nd edn, Klincksieck, 1969). ‘Expansive’ comes from Dowty (cited in McConnell-Ginet,
‘Adverbs and Logical Form’, p. 168), and is also reminiscent of Bach’s ‘expansion’.
18 See Beth Levin, English Verb Classes and Alternations (University of Chicago Press, 1993), for
a detailed description.
19 It is arguably for the same sort of reasons that the agent remains lurking in the background even
after the a g e n t argument-role has been suppressed through passivation. The fact that ‘The ship
was sunk voluntarily / to collect the insurance’ is possible, while ‘The ship sank voluntarily / to
collect the insurance’ is not, is sometimes taken to show that in the passive sentence the a g e n t
argument-role has not been suppressed but remains covertly present at some level of syntactic
structure (Jason Stanley, ‘Making it Articulated’, in Mind and Language 17 (2002), pp. 152–3).
Chomsky argues that the argument-role is present lexically (in the valence of the verb) even
though, syntactically, it is not (Noam Chomsky, ‘Changing Perspectives on Knowledge and Use
of Language’, in Myles Brand and Mike Harnish (eds.), The Representation of Knowledge and
Belief (University of Arizona Press, 1986), pp. 32–5). As far as I am concerned, I favour the
hypothesis that the argument-role is metaphysically implied rather than linguistically realized. A
passive sentence such as ‘The ship was sunk’ results from passivation of the transitive ‘sink’, as in
‘John sank the ship.’ Now the ship plays two roles in (the relations described by) sentences such
as ‘The ship was sunk’ or ‘John sank the ship’: the P-role (p at i e n t or t h i n g a c t e d u p o n ),
and the U-role (u n d e r g o e r o f c h a n g e o f s tat e ). ‘John sank the ship’ means John did
Indexicalism and the Binding Fallacy 109
The variadic functions that increase the valence of the input relation through
the addition of a circumstance to the set of its argument-roles can be represented
by means of an operator (or rather, a family of operators) ‘Circ’. When applied
to an n-place predicate P, ‘Circ’ produces an n + 1 place predicate (‘Circ P’).
There will be as many Circ-operators as there are argument-roles which can be
added to the set of argument-roles of the input relation. There will be a temporal
Circ-operator, a locational Circ-operator, etc., depending on the nature of the
extra argument-role. Which Circ-operator is at issue will be indicated by means
of a subscript. For example, the operator ‘Circlocation ’ contributed by locative
modifiers (such as the prepositional phrase ‘in Paris’) will map, for example,
the e at relation to the e at i n relation by adding a Location argument-role:
Circlocation (Eats (x)) = Eats in (x, l)
As I pointed out, a modifier such as ‘in Paris’ does not merely increase the
valence of the input relation by adding a new argument-role; it also provides
the extra argument needed to fill that argument-role. ‘John eats in Paris’ should
therefore be represented as follows:
Circlocation: Paris (Eats (John)) = Eats in (John, Paris)
Like the prepositional phrase ‘in Paris’, the Circ-operator thus completed does
two things: map the e at relation to the e at i n relation by adding a l o c a -
t i o n argument-role; and supply a particular value (Paris) for that role.20
something to the ship that resulted in the ship’s sinking, where the two tokens of ‘the ship’
correspond to the P-role and the U-role respectively. In ‘The ship sank’, the ship plays only
the U-role. Now nothing can fill the P-role unless there is an agent doing something. Hence
expressions like ‘voluntarily’ or ‘to collect the insurance’, which qualify the manner of acting
or the goal of an agent, can be adjoined to sentences like ‘The ship was sunk’, since the P-role
which occurs in the denoted relation is metaphysically tied to the (linguistically unrealized)
a g e n t role. In contrast, the expression cannot be adjoined to sentences like ‘The ship sank’,
because no action is denoted and no agent is even ‘lurking in the background’. (The explanation
I have just offered is very tentative, of course. There may be decisive syntactic evidence that the
argument-role has not really been suppressed through passivation. But the mere fact that ‘The
ship was sunk to collect the insurance’ is good while ‘The ship sank to collect the insurance’ is
bad is not sufficient to establish such a conclusion, contrary to what Stanley suggests.)
20 According to McConnell-Ginet, who puts forward a similar proposal in her paper ‘Adverbs and
Logical Form’, adverbs such as ‘slowly’ do not contribute an argument filling the extra argument-
role, but they existentially quantify the new argument-role while contributing a property of its
values. If we accept McConnell-Ginet’s idea, ‘John eats slowly’ will be analysed as follows:
Circrate: slow (Eats (John)) = (∃r) (Slow (r) & Eats at rate (John, r)).
That is, there is a rate r which is slow, such that John eats at that rate.
110 Literal Meaning
If we accept (SUP) it follows that the variable which is bound in the complex
sentence has got to be present also, unbound, in the simple sentence ‘It rains.’
Whoever accepts the analysis of adverbial modification in terms of variadic
functions must reject (SUP), however.
According to the variadic analysis, the phrase ‘everywhere I go’ does not
merely contribute what binds the variable, it also contributes the variable it-
self, that is, the extra argument-role for a location. The phrase ‘everywhere
I go’ has a dual role exactly like that of any prepositional phrase. Consider
‘in Paris’. In ‘In Paris it rains’, the prepositional phrase ‘in Paris’ contributes
both (a) a variadic function which adds an extra argument-role to the set of
Indexicalism and the Binding Fallacy 111
argument-roles of the input predicate ‘rain’, and (b) an argument which fills the
role. This duality is quite transparent since the prepositional phrase consists of
two items: a preposition which determines the type of the extra argument-role,
and a name which specifies what fills the role. When the prepositional phrase
is an ‘intransitive preposition’ like ‘here’, it is less obvious that it plays two
semantic roles, but it does so nonetheless. In ‘It rains here’, the locative adverb
‘here’ contributes a variadic function which increases the valence of the ex-
pressed relation, and it also contributes a specific location which fills the extra
argument-role. We find the same duality when the phrase is quantificational
instead of being singular. In ‘Everywhere I go, it rains’, the phrase ‘everywhere
I go’ contributes both the expansive (adicity-increasing) variadic function and
the operator which binds the extra argument-role. From the point of view of
the variadic analysis, therefore, the proper representation of ‘Everywhere I go
it rains’ is:
[For every place l such that I go to l] (in l (it rains))
What the quantifier operates on here is the sub-formula ‘in l (it rains)’, whose
free variable it binds. In that sub-formula we do find a variable for a location.
The simple sentence ‘It rains’ does not correspond to that sub-formula, however,
but to the sub-sub-formula ‘it rains’, which does not contain a free variable for
a location. Stanley’s argument goes through only if we conflate two different
things: the open sentence on which the quantifier operates, and the simple
sentence ‘It rains’ to which the phrase ‘everywhere I go’ has been adjoined. On
the variadic analysis, they are clearly distinguished.
will also not contain a bindable variable, in which case there will be no way to derive
its natural reading . . . Therefore, on the assumption that standard theories of syntactic
ellipsis are correct, there is a bindable variable for quantifier domains present even in
sentences such as (16).22
In this passage, however, Stanley and Szabó commit the Binding Fallacy once
again. Let us see why.
I suggested that the quantified prepositional phrase ‘everywhere I go’ in
‘Everywhere I go, it rains’ has a dual role. It does not merely contribute
what binds the variable, it also contributes the variable itself, that is, the extra
argument-role for a location. From this point of view the proper representation
of ‘Everywhere I go it rains’ is:
[For every place l such that I go to l] (in l (it rains)).
What the quantifier operates on here is the sub-formula ‘in l (it rains)’, whose
free variable it binds. The simple sentence ‘It rains’ does not correspond to that
sub-formula, but to the sub-sub-formula ‘it rains’, which does not contain a free
variable for a location.
The same analysis can be applied to the case of quantifier domain restriction.
On this analysis the second sentence of (17) will be represented as:
[For most classes x such that John has taught x] (in x (John has failed exactly three
Frenchmen)).
What the quantifier operates on here is the sub-formula ‘in x (John has failed ex-
actly three Frenchmen)’, whose free variable it binds. The first sentence of (17)
does not correspond to that sub-formula, however, but to the sub-sub-formula
occurring within the rightmost parentheses. In other words, the quantified prepo-
sitional phrase ‘In most classes he has taught’ contributes not only the quantifier
‘for most classes x such that John has taught x’, it also contributes the variadic
function ‘in x’ with its free variable. All the material that is underlined in the
above formula, including the domain variable, corresponds to what the QPP
contributes. So it is a fallacy to conclude, as Stanley and Szabó do, that ‘if
there is no quantifier domain variable present in the initial sentence in (17),
then there is no way of deriving the natural reading of the second sentence’.
There is an easy way of deriving that reading: one has merely to assume that the
QPP contributes the variable as well as the quantifier which binds it. On this
analysis, the result of copying or reconstructing the first sentence of (17) in
the logical form of the second sentence will not contain a bindable variable,
but that’s fine since the bindable variable is contributed independently by the
QPP.
Of course, it does not follow that quantifier domain restriction is not an in-
stance of saturation. The only thing that follows is that Stanley’s and Szabó’s
argument in favour of an indexicalist treatment in terms of saturation is fal-
lacious and must be disregarded. Contrary to what they claim, they have not
‘demonstrated the existence of quantifier domain variables’.23 It remains an
open question whether quantifier domain restriction is best treated as an in-
stance of enrichment or as an instance of saturation. The Optionality Criterion
does not settle the issue because, as we have seen, it gives no clear result in that
sort of case. In other cases, however, the Optionality Criterion works reason-
ably well; well enough for us to conclude (pace Stanley) that there are cases
in which some ingredient of truth-conditional content results from a pragmatic
process of the optional variety.
23 Ibid., p. 245.
8 Circumstances of evaluation
In this book we have been concerned with the contribution context makes to
truth-conditional content. How essential is that contribution, and how much con-
trolled by linguistic conventions? These are two of the main questions we have
dealt with, in an attempt to revive the debate between Literalism and Contextu-
alism. In this chapter we will consider a fundamental dimension under which
context contributes to truth-conditional content: the so-called ‘circumstance of
evaluation’.
Truth-evaluation (or semantic evaluation more generally) requires not merely
a content to evaluate, but also a ‘circumstance’ against which to evaluate that
content. As Austin once put it, ‘it takes two to make a truth’.1 The circumstance
of evaluation is not an aspect of the content to be evaluated, but an entity with
respect to which that content is evaluated. Still, according to the theory of
situations to be introduced in this chapter, the circumstance of evaluation is an
aspect of content in a broader sense of ‘content’. And that aspect of content is
irreducibly contextual.
8.1 Modality
The notion of circumstance of evaluation is familiar from modal logic. In modal
logic, propositions are evaluated relative to ‘possible worlds’. The possible
worlds are necessary to truth-evaluation, but they are not themselves represented
in the propositions that we evaluate. Thus ‘I am French’ (said by me) is true,
with respect to a world w, iff I am French in w; but the sentence ‘I am French’
only talks about me and the property of being French. The world of evaluation
is not a constituent of the content to be evaluated.
One can bring the world into the content by making the statement more com-
plex. The complex sentence ‘Possibly, I am French’ tells us that in some possible
world I am French. The modal statement I make by uttering that sentence is
about possible worlds, not merely about me and the property of being French.
1 John Austin, ‘Truth’, in his Philosophical Papers (2nd edn, Oxford University Press, 1970),
p. 124 n.
115
116 Literal Meaning
In hybrid logic (a variety of modal logic), one can even make statements ‘refer-
ring’ to specific possible worlds: ‘In w, I am French.’2 But the worlds that are
thus introduced into the content of the complex statement (via modal operators
such as ‘Possibly’ or ‘in w’) are used in evaluating the simple statement ‘I am
French’, that is the statement that is embedded within the modal statement. The
modal statement itself is evaluated with respect to possible worlds, and it shares
with the simple statement the property that the world with respect to which it
is evaluated is not itself represented in the statement under evaluation.3
To appreciate the unarticulated character of the circumstance of evaluation in
the modal framework, it is worth looking at what happens when we (standardly)
translate a modal statement into first-order logic, by explicitly quantifying over
possible worlds. Thus translated ‘Necessarily p’ becomes ‘∀w p(w)’, ‘Possibly
p’ becomes ‘∃w p(w)’, etc. All complete sentences are transformed into pred-
icates (of worlds). A simple categoric statement such as ‘Rain is wet’ will be
represented as ‘p(w)’, where ‘p’ is the proposition that rain is wet transformed
into a predicate of worlds, and ‘w’ is a free variable to which the actual world
is contextually assigned as default value.
The big difference between the modal statement and its standard extensional
translation is that, in the extensional framework, the circumstance of evaluation
(the world) becomes a constituent of content. The contrast between content and
circumstance is lost. This is too bad, for that contrast makes a lot of sense. To
evaluate a sentence, we determine whether the state of affairs it describes obtains
in some ‘reality’ which serves as circumstance of evaluation. But that reality –
the actual world, say – is not itself, or at least doesn’t have to be, among the
constituents of the state of affairs in question, that is, among the entities that are
talked about and articulated in the content of the proposition. The world comes
into the picture for purposes of evaluation, but the thoughts that are evaluated
need not be metaphysically elaborated thoughts about the world. Indeed the
users of the language need not even have the ability to entertain such thoughts.
Only the theorist needs to be able to talk about the world of evaluation, in her
metalanguage. The thoughts that are evaluated ‘concern’ the world, but they
need not be ‘about’ it in the sense in which they are about the entities which
they represent.4
Let us consider a simple language without modal operators or other means
of talking about worlds; let us go further and assume that the users of the lan-
guage don’t possess the reflective abilities necessary for thinking about modal
issues. They entertain only non-modal thoughts such as ‘Rain is wet.’ The
2 See Patrick Blackburn, Maarten de Rijke and Yde Venema, Modal Logic (Cambridge University
Press, 2001), pp. 434–45.
3 The possible worlds which the modal statement talks about are themselves possible with respect to
(that is, accessible from) the world relative to which the modal statement happens to be evaluated.
4 See Perry, ‘Thought Without Representation’, for the distinction between ‘concerning’ and ‘being
about’.
Circumstances of evaluation 117
possible-worlds semanticist who studies their language will still need to think
and talk about the possible worlds relative to which the sentences of the lan-
guage are evaluated; but, contrary to what the standard extensional translation
suggests, mention of the possible worlds in question will be confined to the
theorist’s metalanguage.
Now suppose the users of the object-language become sophisticated and start
thinking about metaphysical issues. Suppose they come to talk and think about
what is actually the case as opposed to what might be the case. Such modal talk
can be formally represented in two ways, as we have seen: by using sentence
operators, or by explicitly quantifying world variables in the object-language. If
we use the modal framework and introduce modal operators such as ‘actually’
or ‘possibly’, nothing will be changed for the fragment of the language that does
not involve those operators. The sentence ‘Rain is wet’ will still be a simple,
modally innocent sentence. The language will simply have been enriched by the
introduction of new resources enabling us to construct more complex sentences.
But if we use the standard extensional framework and represent modal sentences
(‘It might be that . . .’, ‘Actually . . .’) by means of explicit quantification over
possible worlds, as suggested above, then, unless special precautions are taken
to avoid that consequence, a change of language takes place, not merely an
enrichment. In the new language, all sentences (including simple sentences)
now contain a hidden argument-place for a world. Modal innocence is lost.
I think this move is (almost) as damaging as the previous one – the ascrip-
tion of thought and talk about possible worlds to modally innocent subjects.
Even if the users of the language are sophisticated enough and can think about
modal issues, it is misleading to suggest that they always think and talk about
such issues even when they entertain simple thoughts or utter simple sentences
such as ‘Rain is wet.’ By forcing us to construe for example the assertion that
rain is wet as involving a covert argument-place which the actual world fills,
the extensional translation blurs the cognitively important distinction between
the simple, modally innocent assertion ‘Rain is wet’ and the modal assertion
‘Actually, rain is wet.’ To maintain that distinction, we have to see modal
sentences as constructed from simple sentences by the application of modal
operators to them. In this way we can analyse the ability to use and understand
modal sentences as resting on two distinct abilities: the ability to use and un-
derstand simple sentences; and the ability to imagine other possible worlds and
to contrast the actual world with them. The first ability is independent of the
second: we can use and understand simple sentences (for example, ‘Rain is
wet’) even if we lack the ability to think thoughts about the actual world (in
Perry’s sense of ‘about’).5
5 The important thing, I said, is to see modal sentences as constructed from simple sentences by
means of operators. Now this is something we can do even if we want to represent modal talk
extensionally. The apparatus of variadic functions enables us to do just that. In §7.6 I analysed
118 Literal Meaning
‘Everywhere I go it rains’ as resulting from the application of a locative variadic operator to the
sentence ‘It rains.’ That operator does two things. First, it modifies the adicity of the predicate in
the sentence it applies to: it adds an extra argument-place for a location, which can be represented
by a free variable. Second, it introduces a restricted quantifier which binds that variable. The
operator can be paraphrased as ‘for every location l such that I go to l, in l it is the case that’.
‘Necessarily it rains’ can be represented in the same hybrid way, by applying to the sentence ‘It
rains’ a sentence operator which can be rendered as: ‘for every world w, in w it is the case that’.
Since the variable w is introduced by the variadic operator, we don’t have to treat the emergence
of modalities as a radical change in the language, but simply as an enrichment of it; an enrichment
which does not affect the simple (non-modal) sentences, hence preserves modal innocence.
6 In ‘Structural Analogies Between Tenses and Pronouns’, Barbara Partee says that examples like
‘I did not turn off the stove’ (in which reference is made to a specific time) speak against a
treatment in terms of operators, because modal operators can’t capture the referential nature of
(some uses of) tenses. But the referential/quantificational issue is orthogonal to the question,
whether or not we should use operators. Even if standard modal operators are quantificational
rather than referential, nothing prevents the introduction of ‘referential’ operators in the modal
framework. See Arthur Prior, Past, Present, and Future (Clarendon Press, 1967) and ‘Now’ in
Noûs 1(1968), 101–19; and Patrick Blackburn, ‘Tense, Temporal Reference and Tense Logic’,
in Journal of Semantics 11(1994), 83–101.
7 There is a third option: tenses can be represented as temporal predicates of events. If we like
Davidson’s analysis of adverbial modification better than the variadic analysis, that is a natural
move to make.
Circumstances of evaluation 119
advance is made by our acquiring a general grasp of the past and future tenses. That is
to say, to understand ‘was warm’ or ‘will be warm’, we apply to our prior understanding
of what is meant by saying that something is warm our general comprehension of what
it is to speak of how things were or will be at another time. In so doing, we are in effect
treating the tenses (and other indications of time) as operators applied to sentences in
the present tense of which we have previously acquired an understanding, just as the
tense-logical semantics treats them. We could not learn the language in any other way.8
8 Michael Dummett, ‘Existence, Possibility, and Time’ (typescript, Oxford University), pp. 16–17.
120 Literal Meaning
of an event, we may say, is simply the occurrence of the event, and that is simply the
event itself. But every complete tensed sentence characterises the time of utterance in
some way or other, and other times only through their relation to that one.9
To sum up, the time of utterance is not represented, it does not feature in the
content of tensed sentences; it only comes into the picture as the circumstance
with respect to which the content of a tensed sentence is evaluated.
Another objection, voiced by Evans,10 concerns the fact that a tensed sentence
like ‘It is hot’, ‘It has been hot’ or ‘It will be hot’ is not evaluable as true or false,
unless we are given a particular time. In the absence of a time specification,
the sentence is only ‘true-at’ certain times and ‘false-at’ others. Such a sentence,
therefore, is semantically incomplete by Frege’s lights:
A thought is not true at one time and false at another, but it is either true or false, tertium
non datur. The false appearance that a thought can be true at one time and false at
another arises from an incomplete expression. A complete proposition or expression of
a thought must also contain a time datum.11
As Evans points out, the problem of semantic incompleteness does not arise
in the modal case. Even if a thought is said to be ‘true at’ one world and
‘false at’ another, as in modal logic, this does not prevent it from being true
(or false) tout court. It is true tout court iff it is true-at the actual world. But
the ‘thought’ that it is hot cannot be evaluated as true or false tout court. In
the absence of a contextually supplied time it can only be ascribed relative,
‘truth-at’-conditions. Only a particular, dated utterance of such a sentence can
be endowed with genuine truth-conditions. What this shows is that the time of
utterance is part of the (complete) content of the utterance;12 hence it cannot
be expelled out of the content and treated like the world of evaluation. So the
objection goes.
According to Dummett, Evans’s objection to Prior is based on a misunder-
standing. Prior was concerned only with sentence-types and their contents. The
content of a sentence-type is a function from times to truth-values, hence a
sentence-type has only relative truth-conditions: it is true at some times and
false at other times. To introduce a notion of absolute truth, one thing we can
say (though not, according to Dummett, what Prior himself would say)13 is
9 Arthur Prior, ‘Egocentric Logic’, in Arthur Prior and Kit Fine, Worlds, Times and Selves
(Duckworth, 1977), p. 30.
10 See Gareth Evans, ‘Does Tense Logic Rest on a Mistake?’, in his Collected Papers (Clarendon
Press, 1995), pp. 350–3.
11 Gottlob Frege, Kleine Schriften (Olms, 1967), p. 338, quoted in Evans, ‘Does Tense Logic Rest
on a Mistake?’, p. 350.
12 Or, in a Fregean framework, part of the expression of such a content.
13 ‘The simplest way to introduce a notion of absolute truth’, Dummett writes, ‘is to follow the
analogy with possible worlds semantics and stipulate a type sentence to be true simpliciter
just in case it is true-now. Tense-logic, in the hands of its inventor, could be regarded, without
violation of its principles, as a semantics exclusively of statements uttered at one particular time’
(‘Existence, Possibility, and Time’, p. 19).
Circumstances of evaluation 121
that, when a sentence is uttered, the function which is its content is applied to
some contextually provided time (typically, the time of utterance). The time in
question serves as circumstance of evaluation for the utterance: the utterance
is true tout court iff the sentence is ‘true-at’ the contextually provided time. As
Dummett points out,
The variable truth-value and the absolute truth-value attach to different things; it is the
type sentence that is true at one time, false at another, but the utterance that is true or
false simpliciter.14
Since there are two distinct levels, corresponding to the sentence-type and
the utterance, there is no harm in taking the utterance to possess a ‘content’ also
(contentu ), distinct from that of the sentence (contents ). For example, we can
treat the utterance as expressing a structured proposition consisting of (i) the
contextually provided time as subject, and (ii) the content of the sentence-type,
predicated of that time. But if we do so, we must acknowledge the unarticu-
lated nature of the ‘subject’ in the contentu of tensed utterances. As Prior says,
‘tensed propositions are understood as directly or indirectly characterizing the
unmentioned time of utterance’.15 Hence there is a trade-off: if we want to
restrict ourselves to what is linguistically articulated, we must focus on the
contents , which is ‘semantically incomplete’ by Frege’s lights – it corresponds
to the content of a predicate rather than to that of a complete sentence in a log-
ically perfect language. If, following Frege, we want to focus on the complete
content of the utterance, that which makes it truth-evaluable in absolute terms,
we must acknowledge the role played in that content (contentu ) by unarticu-
lated constituents corresponding to the circumstances in which the contents is
evaluated.
8.3 Situations
Let us take stock. For purposes of semantic evaluation we need a circumstance
as well as a content. Even Frege, who was unconcerned with modalities and
thought of the actual world as the only world there is, was aware of that fact.
He took fictional sentences to be unevaluable, for the following reason: since
the author of a fictional statement does not attempt to characterize the actual
world, we are given a content without any circumstance of evaluation for it. The
obvious conclusion to draw from Frege’s remarks on fiction is that, to get a truth-
value, a content is not sufficient; we need to connect that content with the actual
world, via the assertive force of the utterance, in virtue of which the content is
presented as characterizing that world. Frege was aware not only that we need
a circumstance in addition to a content, but also that the circumstance is not,
and cannot be, an aspect of the content articulated in the sentence. If a sentence
lacks the force of a serious assertion, because the speaker does not attempt
to characterize the actual world but is engaged in a different enterprise (for
example poetry), making the content of the sentence more complex by means
of an additional phrase such as ‘it is true that’ will not change the situation.
Whether or not an utterance is serious and characterizes the actual world is a
pragmatic matter – a matter of ‘force’, not a matter of content (in the narrow
sense of ‘content’).16
Once it is admitted that we need a circumstance over and above the content
to be evaluated, we can part from Frege and, following Prior, tolerate contents
that are not ‘semantically complete’ in Frege’s sense, that is, endowed with
absolute truth-conditions. We can, because the circumstance is there which en-
ables the content to be suitably completed. Thus the content of tensed sentences
is semantically incomplete, yet the circumstance (the time) relative to which
such a sentence is evaluated is sufficient to complete it. It follows that we must
distinguish two levels of content. The content we evaluate with respect to the
circumstance is the contents ; it may, but need not be, semantically complete
by Frege’s lights. What is semantically complete in any case is the contentu . It
consists of the contents and the circumstance with respect to which the contents
is evaluated.
Situation theory as I understand it17 follows those ideas to their consequences.
It generalizes and systematizes them, in two main directions:
(1) There is no reason why only times and worlds should be accepted as features
of the circumstance of evaluation. Why not also, for example, locations? If I
say ‘It’s raining’, the location is unarticulated, but it is relevant qua feature of
the circumstance of evaluation: what I say (or think) is true iff it’s raining at the
contextually provided location. Why not also consider the agent of the speech
act (the speaker) or the thought act (the thinker) as (part of) the circumstance of
evaluation, to handle the cases in which the content to be evaluated is a property
of agents which the speaker or thinker self-attributes?18 Why not extend the
notion also to ordinary objects? If, talking about my car, the mechanic tells
me, ‘The carburettor is in good condition but there is a problem with the front
16 See Gottlob Frege, ‘Thought’, in Michael Beaney (ed.), The Frege Reader (Blackwell, 1997),
p. 330.
17 By ‘situation theory’ here I do not mean the official doctrine expounded in Barwise’s and Perry’s
Situations and Attitudes (MIT Press, 1983), but a later doctrine, influenced by Perry’s ‘Thought
Without Representation’ and centred around the notion of Austinian proposition. See Jon Barwise
and John Etchemendy, The Liar: An Essay on Truth and Circularity (Oxford University Press,
1987); Jon Barwise, ‘Situations, Facts, and True Propositions’, in his collection The Situation in
Logic (CSLI Publications, 1989), 221–54; François Recanati, ‘The Dynamics of Situations’, in
European Review of Philosophy 2 (1997), 41–75, ‘Situations and the Structure of Content’, and
Oratio Obliqua, Oratio Recta. An Essay on Metarepresentation (MIT Press, 2000), chapters 5
to 7; and Jérôme Dokic, ‘Steps Toward a Theory of Situated Representations’ (typescript, Institut
Jean Nicod).
18 See David Lewis, ‘Attitudes de dicto and de se’, reprinted (with a postscript) in his Philosophical
Papers vol. 1 (Oxford University Press, 1983), pp. 133–59.
Circumstances of evaluation 123
I say: ‘Claire has a good hand.’ What I say is true, iff Claire has a good hand in
the poker game I am watching (at the time of utterance). But suppose I made a
mistake and Claire is not among the players in that game. Suppose further that,
by coincidence, she happens to be playing poker in some other part of town
and has a good hand there. Still, my utterance is not intuitively true, because
the situation it concerns (the poker game I am watching) is not one in which
Claire has a good hand. But we can say that the sentence is true, or at least true
at the time of utterance: for it says that Claire has a good hand, and Claire has
a good hand (somewhere). The unarticulated constituent which distinguishes
the contentu from the contents makes all the difference here, and it accounts for
our intuitive classification of the utterance as non-true.
possibly more than one contextual domain, in a given sentence. The sentence
‘Every sailor waved to every sailor’ can be used to say that every sailor on the
ship waved to every sailor on the shore.22 Since we can have only one circum-
stance of evaluation for that sentence, how can we deal with this multiplicity
of contextual domains? This is also the objection raised by Scott Soames to the
situation-theoretic treatment of incomplete definite descriptions.23 As I pointed
out in the third section of ‘Domains of Discourse’,24 the answer to this difficulty
consists in multiplying situations of evaluation. This can be done by extending
the notion of a circumstance of evaluation to sub-sentential expressions. Thus,
following Kuroda, I have argued that every predicate requires a situation of
evaluation.25 Let me briefly sketch a treatment of quantifier domain restriction
within that framework.
In a standard conception, predicates denote sets of objects, and quantifiers
denote relations between sets of objects. In ‘Every F is G’, ‘every’ denotes
that relation which holds between the set of Fs and the set of Gs just in case
the former is included in the latter. Now predicates denote sets of objects only
relative to situations. It follows that, before we can evaluate a quantificational
statement, we must evaluate the predicates with respect to situations so as to
determine the sets of objects which serve as arguments to the quantifier.
When I say that predicates denote sets of objects only relative to situations,
I intend this as a rather trivial point. Some objects are red in a given situation,
which may no longer be red in a different (say, temporally posterior) situation.
So the set of red objects is variable and depends upon the situation at stake,
even if we do not vary the domain from one situation to the next. There may be
properties which stick to their objects in the sense that, if an object has them in
a situation, it must have them in any situation to the domain of which that object
belongs. Even if there are such sticking properties, still the predicates which
correspond to them will possibly denote different sets of objects in different
situations because the domain of objects itself can vary from one situation to
the next.
So predicates require situations for their evaluation. The same thing holds for
sentences, as we have seen. Whether a sentence is true or false depends upon how
22 This example comes from Jason Stanley and Tim Williamson, ‘Quantifiers and Context-
Dependence’, in Analysis 55 (1995), 291–5.
23 Scott Soames, ‘Incomplete Definite Descriptions’, in Notre Dame Journal of Formal Logic 27
(1986), 349–75.
24 François Recanati, ‘Domains of Discourse’, in Linguistics and Philosophy 19 (1996), pp. 445–
75.
25 Yuki Kuroda, ‘Indexed Predicate Calculus’, in Journal of Semantics 1 (1982), 43–59; François
Recanati, ‘Contextual Dependence and Definite Descriptions’, in Proceedings of the Aristotelian
Society 87 (1987), 57–73, and ‘Descriptions and Situations’, in Anne Bezuidenhout and Marga
Reimer (eds.), Descriptions and Beyond (Oxford University Press, forthcoming).
Circumstances of evaluation 127
things are in the relevant situation of evaluation. Now the main predicate in that
sentence – the predicate which corresponds to the topmost verb-phrase – will
always be evaluated with respect to the situation of evaluation for the sentence in
question. But the other predicates in the sentence will possibly involve distinct
situations of evaluation (‘resource situations’, in the terminology of Barwise
and Perry, Situations and Attitudes).
Consider the sentence: ‘Every student laughs.’ There is no possibility of a
divergence between the situation with respect to which the sentence is evaluated
and the situation with respect to which the main predicate, ‘laughs’, is evaluated.
That means that, if the sentence is evaluated with respect to some situation s,
the set of laughers which serves as second argument to the quantifier ‘every’
will be the set of laughers-in-s. In contrast, the predicate which occurs as part
of the noun-phrase, ‘student’, can be evaluated with respect to any situation: it
may be the situation of evaluation s (in which case every student-in-s is said
to be among the laughers-in-s), it may be the situation of utterance c (in which
case every student-in-c is said to be among the laughers-in-s), or it may be any
auxiliary situation which happens to be sufficiently salient for either linguistic
or extra-linguistic reasons.
In this framework, we can handle the problem of quantifier domain restriction
(and the analogous problem of incomplete definite descriptions) via the provi-
sion of a situation of evaluation, or of a sequence of such situations. Again, this
does not settle the issue, whether the provision of a situation for the evaluation
of a given predicate is an instance of enrichment or of saturation. It’s an in-
stance of saturation if, like Stanley and Szabó, we take a situation variable (or
something like that) to be associated with every common noun in logical form.
But we may also consider that the provision of a situation of evaluation is not
the assignment of a value to a variable in logical form, but an optional process
which may or may not take place. Whenever that process does not take place, the
situation of evaluation for the predicate can only be the situation of evaluation
for the sentence. That means that if the sentence is semantically complete and
we evaluate it directly relative to the actual world, the set of objects denoted by
the predicate will be the set of objects which satisfy the predicate in the actual
world.
8.6 Conclusion
As we have seen at the beginning of this chapter, the possible worlds relative to
which, in modal logic, sentences are evaluated are not aspects of the contents
to be evaluated, but remain external to those contents. The same thing holds
when time is the relevant circumstance of evaluation. Yet there is a difference
between time and modality: the type of content that is evaluated with respect
128 Literal Meaning
to a time in tense logic is not semantically complete in the sense in which the
content which is evaluated with respect to a possible world is. This difference
was emphasized by Evans in his attack on tense logic. But the proper response
to make, on behalf of tense logic, is the drawing of a distinction between two
levels of content. The content which a sentence articulates is one thing; the
content of an utterance of that sentence, with respect to a given circumstance of
evaluation, is another thing. Only the second, richer type of content is a legiti-
mate bearer of truth-value, by Fregean lights. It follows that the truth-conditions
of a tensed utterance crucially involve the circumstance of evaluation, which is
not an aspect of the content articulated in the sentence. Hence there is a funda-
mental aspect of truth-conditional content that is not articulated in the uttered
sentence.
The theory of situations I briefly introduced generalizes the lessons from tense
logic. In this theory every utterance articulates a certain content and expresses a
richer type of content (an ‘Austinian proposition’) consisting of the articulated
content and the situation with respect to which that content is (intended to be)
evaluated. The complete, Austinian proposition expressed by an utterance is
true iff the situation of evaluation supports the articulated content. That content
is typically not a classical proposition but a relativized proposition – a function
from rich circumstances to truth-values.
In this framework there is an ‘unarticulated constituent’ in the (complete)
content of every utterance, namely the circumstance of evaluation. Admittedly,
there are different types of unarticulated constituent.26 In many cases a cir-
cumstantial constituent counts as ‘unarticulated’ only in a weak sense. That is
so whenever the constituent, without being explicitly represented in the articu-
lated content, is nevertheless mandatory because that content is not semantically
complete by Frege’s lights. Thus the time of evaluation (that is, the time relative
to which an utterance is evaluated) is unarticulated because, as Prior says, it is
‘unmentioned’ in the utterance, yet its contextual provision results from a form
of saturation: such a constituent must be provided in order to get a complete
proposition, that is, a function from possible worlds to truth-values. In situation
theory, however, the necessity to provide the relevant circumstantial constituent
is exactly like the necessity to provide a possible world in order to determine
a truth-value. In general, contents need circumstances for their evaluation.
Once this is acknowledged, the door is open to relativized contents of all sorts,
and the privilege given to classical propositions (functions from possible worlds
to truth-values) over relativized propositions (functions from rich circumstances
to truth-values) looks somewhat arbitrary.27 Be that as it may, a distinction must
26 See footnote 4, p. 99.
27 Classical propositions fulfil two functions: they are the bearers of absolute truth-values, and they
are the object of the attitudes. In the situation theoretic framework, the Austinian proposition
Circumstances of evaluation 129
be drawn in all cases between the articulated content (contents ) and the complete
content of the utterance (contentu ); and the irreducibly contextual nature of the
circumstance that features in the contentu must be acknowledged. Insofar as
this position generalizes the context-dependence of truth-conditional content,
it can be categorized as ‘contextualist’.
Still, one might argue, the theory I have sketched posits a level of content that
is very close to the linguistic meaning of the sentence: the contents . Insofar as it
posits such a level, the theory of situations sounds more similar to the Syncretic
View than to Contextualism or even Quasi-Contextualism. I find this reaction
natural, yet I maintain that the theory of situations stands on the contextualist
side. I concede that there are similarities with the Syncretic View, but they are
more superficial than they seem.
The main analogy consists in the fact that, in both cases, we find two levels
of content: the proposition expressed by the sentence, and a richer proposition
containing unarticulated constituents.28 Taken literally, ‘You’re not going to
die’ is true iff the addressee is not going to die. This corresponds to the minimal
proposition of the Syncretic View (what is strictly and literally said), and to the
contents of situation theory. That content yields falsity in all possible worlds in
which the addressee is not immortal. So the utterance counts as false (‘literally
false’) if we evaluate the contents directly with respect to the actual world. If we
take the intended situation of evaluation into account and evaluate the complete
Austinian proposition (contentu ), the utterance counts as true iff the addressee
is not going to die in that particular situation (resulting from the cut). This
truth-condition corresponds to the intuitive truth-conditions of the Syncretic
View.
Despite this convergence between the two views, there are important differ-
ences. The articulated content (right-hand-side of the Austinian proposition) is
unlike the minimal proposition of the syncretist in two respects:
(1) As we have seen, the articulated content of the situation theorist need not
be a complete proposition; it may be a relativized proposition. In contrast, the
minimal proposition must be a complete proposition, because the ‘minimalist’
criterion that is used in individuating it involves truth-evaluability.29
fulfils the first function, and the articulated content (relativized proposition) fulfils the second
function, for which classical propositions are in any case ill-suited. So it is unclear that we need
classical propositions.
28 I stressed that analogy in ‘Situations and the Structure of Content’, pp. 121–2.
29 Note, however, that some syncretists use a different criterion, and can therefore accommodate
relativized propositions. Thus for Bach (who uses the ‘IQ test’ as criterion), what is literally said
need not be a complete proposition, it may be a ‘propositional radical’. Soames has recently
come to endorse a similar view. See his ‘Naming and Asserting’, in Zoltán Szabó (ed.), Semantics
and Pragmatics (Oxford University Press, forthcoming).
130 Literal Meaning
(2) More important in the context of the present discussion, there is abso-
lutely no reason why the articulated content (contents ) of the situation theorist
should not be affected by primary pragmatic processes of the optional variety.30
Typically, the situation that is contextually provided for the evaluation of an ut-
terance will contribute to shaping the content that is evaluated with respect to
that situation: the meaning of words will naturally be enriched and adjusted to
fit the intended situation of evaluation. So the articulated content of the situation
theorist is definitely not the minimal proposition posited by the Syncretic View.
30 Hence it would be misleading to keep describing this content as ‘the content of the type sentence’,
to use Dummett’s words. The contents is as much a property of the utterance as the contentu . The
contents is the articulated content of the utterance, while the contentu is the complete content
of the utterance (including the intended circumstance of evaluation).
9 Contextualism: how far can we go?
131
132 Literal Meaning
I agree with Cohen that the semantics of natural language is not insulationist.
As I have been at pains to emphasize, the meaning of the whole is not constructed
in a purely bottom-up manner from the meanings of the parts. The meaning of the
whole is influenced by top-down, pragmatic factors, and through the meaning
of the whole the meanings of the parts are also affected. So we need a more
‘interactionist’ or even ‘Gestaltist’ approach to compositionality.7 But I think
Cohen takes too narrow a view of the sort of interaction that sense modulation
results from.
Let us go back to the example discussed by Michael Pelczar:
In these sentences:
‘There was a light breeze from the south.’
‘This is a light package.’
‘The fire won’t last with such light fuel.’
– ‘light’ expresses a content having to do with intensity, weight, and density, respectively.
And this clearly results from the fact that in the first sentence ‘light’ is used to speak of
a breeze, whereas in the second it is applied to describe a package, and in the third, to
some fuel for a fire.8
According to Cohen, Ross and many others, the sense of ‘light’ is modulated
through its interaction with the meaning of the head noun (‘breeze’, ‘package’,
‘fuel’).9 I think this is not quite right. The meaning of the head-noun is not what
matters most. It is, as Pelczar emphasizes, what the word ‘is used to speak of’,
what it ‘is applied to describe’, that matters. Nothing prevents us from using
‘light’, in the same phrase ‘light fuel’, to characterize the weight of the fuel
rather than its density, if this is what is relevant to us in the context at hand.
We can even imagine a context in which ‘light’, as applied to a package, would
characterize the intensity of the sound it emits.
To take another example from Pelczar’s study, the word ‘get’ takes on dif-
ferent senses – denotes different relations – depending on what fills the second
argument-place of the relation. Thus ‘Smith has got the virus’ means that he has
contracted the virus, while in ‘I need to get some eggs’ ‘get’ has what Pelczar
calls the acquisitional sense. Yet, as Pelczar points out, we can easily imagine
a context in which ‘Smith has got the virus’ means that Smith has acquired,
that is, successfully collected a sample of, the virus. The same remarks can be
made with respect to the word ‘cut’, whose meaning is modulated differently
in ‘John cut the grass’ and ‘John cut the cake.’ As Searle points out, if I ask
John to cut the grass, and he cuts it the way one cuts a cake (or if I ask him to
cut the cake, and he cuts it the way one cuts grass), he has not complied with
my request; he has not done what I asked him to do. Still, we can imagine a
context in which in asking John to cut the grass, I would mean that he is to cut
it the way one cuts a cake:
Suppose you and I run a sod farm where we sell strips of grass turf to people who want
a lawn in a hurry . . . Suppose I say to you, ‘Cut half an acre of grass for this customer’;
I might mean not that you should mow it, but that you should slice it into strips as you
could cut a cake or a loaf of bread.10
So it’s not just the interaction of the meaning of the word with the linguistic
context that is relevant. The really important factor is the discourse topic. The
‘internal’ interaction of word-meanings within the same sentence or discourse
is only a reflection of the more general phenomenon: the interaction between
word meaning and the situation the words are used to talk about.
9 This is what Lahav calls the ‘noun-dependence’ of adjectives and verbs – a feature that, according
to him, threatens the Principle of Compositionality.
10 Searle, ‘The Background of Meaning’, pp. 224–5.
134 Literal Meaning
11 This can be established by appealing to examples like ‘Mary is wearing a light coat; so is Jane.’
For that to be true, Jane’s coat must be light in the same modulated sense as that in which Mary’s
coat is said to be light. (I have borrowed this example from Alan Cruse, Meaning in Language:
An Introduction to Semantics and Pragmatics (Oxford University Press, 2000), p. 106.)
12 See for example ‘Making Sense of Nonce Sense’ and ‘Understanding Old Words with New
Meanings’, two papers published by Herb Clark in 1983 (the second one co-authored by Richard
Gerrig) and reprinted respectively as chapter 10 and chapter 11 in his Arenas of Language Use
(The University of Chicago Press, 1992). See also Jay Atlas, Philosophy without Ambiguity
(Clarendon Press, 1989), pp. 29–31, and Pustejovsky, The Generative Lexicon, chapter 4.
13 See Catherine Fuchs, ‘Polysémie, Interprétation et Typicalité’, in Danièle Dubois (ed.),
Sémantique et Cognition: Catégories, Prototypes, Typicalité (Editions du CNRS, 1991),
Contextualism: how far can we go? 135
Can we say at least that the phenomenon of modulation only concerns a re-
stricted class of words, namely those that are ‘polysemous’ and possess a family
of related senses? No. First, as Cohen points out, polysemy is the norm rather
than the exception.14 To be sure, not all polysemous words are as polysemous
as ultrafrequent words such as ‘get’. Polysemy comes in grades, and it could
be argued that modulation is associated with a high degree of polysemy: highly
polysemous words, it may be said, have an abstract core meaning which calls
for elaboration or modulation, and from which the various senses of the poly-
semous expression are generated.15 This does not entail that non-polysemous
words are not subject to modulation, however. Polysemy arguably proceeds
from modulation (since the array of senses of a polysemous expression re-
flects conventionalized patterns of modulation), not the other way round.16 So
the proper response to make is that modulation can affect any ordinary word,
whether or not it is (highly) polysemous.
A paradigm example of modulation is the use of colour terms such as ‘red’.17
The conditions of application for ‘red’ depend upon (inter alia) what sort of thing
the colour adjective is applied to. As Lahav puts it, the conditions that a table has
to meet in order to be describable by ‘red table’ are not the same conditions that
a house or a book or a bird has to meet in order to be describable by ‘red house’
or ‘red book’ or ‘red bird’.18 Yet ‘red’ is not especially polysemous; it is just an
pp. 165–7, on French ‘pouvoir’; Sue Atkins, ‘The Contribution of Lexicography’, in Madeleine
Bates and Ralph Weischedel (eds.), Challenges in Natural Language Processing (Cambridge
University Press, 1993), pp. 39–43, on ‘safety’; and Richard Grandy, ‘Understanding and the
Principle of Compositionality’, in Philosophical Perspectives 4 (1990), p. 571, on ‘prime’.
14 Jonathan Cohen, ‘A Problem About Ambiguity in Truth-Conditional Semantics’, in Analysis 45
(1985), p. 131.
15 See Alfonso Caramazza and Ellen Grober, ‘Polysemy and the Structure of the Subjective Lex-
icon’, in Clea Rameh (ed.), Semantics: Theory and Application (Georgetown University Press,
1976), 181–207.
16 ‘Ce que l’on appelle la polysémie n’est que la somme institutionnalisée, si l’on peut dire, de
ces valeurs contextuelles, toujours instantanées, aptes continuellement à s’enrichir, à disparaı̂tre,
bref, sans permanence, sans valeur constante’ (Emile Benveniste, ‘La Forme et le Sens dans le
Langage’, in his Problèmes de Linguistique Générale II (Gallimard, 1974), p. 227).
17 See for example Charles Travis, The True and the False (Benjamins, 1981), p. 16, and Lahav,
‘Against Compositionality’, p. 264.
18 ‘For a bird to be red (in the normal case), it should have most of the surface of its body red,
though not its beak, legs, eyes, and of course its inner organs. Furthermore, the red color should
be the bird’s natural color, since we normally regard a bird as being ‘really’ red even if it is
painted white all over. A kitchen table, on the other hand, is red even if it is only painted red, and
even if its ‘natural’ color underneath the paint is, say, white. Morever, for a table to be red only
its upper surface needs to be red, but not necessarily its legs and its bottom surface. Similarly,
a red apple, as Quine pointed out, needs to be red only on the outside, but a red hat needs to
be red only in its external upper surface, a red crystal is red both inside and outside, and a red
watermelon is red only inside. For a book to be red is for its cover but not necessarily for its
inner pages to be mostly red, while for a newspaper to be red is for all of its pages to be red. For
a house to be red is for its outside walls, but not necessarily its roof (and windows and door)
to be mostly red, while a red car must be red in its external surface including its roof (but not
136 Literal Meaning
its windows, wheels, bumper, etc.). A red star only needs to appear red from the earth, a red
glaze needs to be red only after it is fired, and a red mist or a red powder are red not simply
inside or outside. A red pen need not even have any red part (the ink may turn red only when in
contact with the paper). In short, what counts for one type of thing to be red is not what counts
for another’ (Lahav, ‘Against Compositionality’, p. 264).
19 Cohen, ‘Ambiguity in Truth-Conditional Semantics’, p. 132.
Contextualism: how far can we go? 137
The pragmatic process that enriches the meaning of the first utterance so as to
convey both a sense of temporal order (giving to ‘and’ the sense of ‘and then’)
and a notion of the instrument used in opening the door (giving to ‘opened the
door’ the sense of ‘opened the door with the key’) – that process might also
not take place. As Grice emphasized,20 such pragmatic suggestions are always
cancellable, explicitly or contextually.21 Once the pragmatic suggestion has
been cancelled, what the words contribute to truth-conditional content is their
bare linguistic senses. The same thing can be said about the second example.
‘Lion’ can be understood in the representational sense: the ‘lion’ that is talked
about can be a representation (for example, a statue) of a lion, not a real lion.
That reading results from a primary pragmatic process that can be variously
described.22 However we describe it, its optional character is obvious: nothing
prevents that sentence from being understood literally, as talking about a real
lion.
The three other views I want to discuss are all contextualist. They all ascribe
to modulation a form of necessity or ineliminability, to the point of blurring
the mandatory/optional contrast. Discussing them will enable us to distinguish
more or less radical varieties of Contextualism. The last view I shall consider
corresponds to the most radical form: Meaning Eliminativism.
Like SO, the Pragmatic Composition view (PC) accepts that the literal, in-
put sense undergoing modulation could, in a suitable context, be the expressed
sense. So it construes the pragmatic process of modulation as optional. But it
construes it as optional only with respect to the word whose sense is modu-
lated. If we consider not words in isolation, but the complex expressions in
which they occur, we see that the primary pragmatic processes of modulation
are not always contingent and dispensable, but often essential. Even though
the linguistic meaning of a given word (or the semantic value that results from
saturation) could be the expressed sense, still the process of semantic composi-
tion, that is, the putting together of that sense with the semantic values for other
expressions, cannot proceed unless appropriate adjustments take place so as to
make the parts fit together within an appropriate whole. On this view words
have meanings which could go directly into the interpretation, without modu-
lation, but it is the composition process that forces modulation to take place,
or at least invites it: often the meanings of individual words do not cohere by
themselves, and can be fitted together only by undergoing a process of mutual
adjustment.
Remember the examples I gave in chapter 2. If we take it as axiomatic that
only sounds can be heard, then, in ‘I hear the piano’, either the sense of ‘hear’ or
that of ‘the piano’ must be modulated for the sentence to make sense (§2.6).23
Or remember the Gary Cooper example, ‘disarm the fricassee’, or Searle’s
example, ‘Cut the sun’, from §6.3. To make sense of such sentences we must
imagine (or retrieve from memory) a possible scenario in which the semantic
values of the words fit together.24 That imaginative exercise involves elaborating
what the meaning of the words gives us – going beyond that linguistic meaning
and, for example, interpreting ‘disarm’ in the specific sense of: ‘take the gun
out of’ or ‘remove the gun from’.
Or consider the paradigmatic example I gave in the last section: the adjective
‘red’. There is no particular incompleteness in the linguistic meaning of that
23 Note that there may be no fact of the matter as to which particular word or word-meaning
undergoes modulation, in examples like those of chapter 2. This is due to the holistic character
of modulation.
24 See Rumelhart, ‘Problems with Literal Meanings’, p. 77. See also Anthony Sandford and Simon
Garrod, ‘The Role of Scenario Mapping in Text Comprehension’, in Discourse Processes 26
(1998), 159–90.
Contextualism: how far can we go? 139
word – it means something like ‘being of the colour red’ or ‘having the colour
red’. Insofar as ‘red’ refers to a specific colour (and it does) this, it seems,
expresses a definite property – a property that could, in principle, go into the
interpretation of a sentence in which the adjective ‘red’ occurs. (For example:
‘Imagine a red surface.’) But in most cases the following question will arise:
what is it for the thing talked about to count as having that colour? Unless that
question is answered, the utterance ascribing redness to the thing talked about
(John’s car, say) will not be truth-evaluable. It is not enough to know the colour
that is in question (red) and the thing to which that colour is ascribed (John’s
car). To fix the utterance’s truth-conditions, we need to know something more –
something which the meanings of the words do not and cannot give us: we
need to know what it is for that thing (or for that sort of thing) to count as
being that colour. What is it for a car, a bird, a house, a pen, or a pair of shoes
to count as red? To answer such questions, we need to appeal to background
assumptions and world knowledge. Linguistic competence does not suffice:
pragmatic fine-tuning is called for.
To sum up, on the view (PC) under discussion, even if the semantic value of
a word is fixed by language (and context, if saturation is necessary), composing
it with the semantic values for other words often requires help from above. It
is semantic composition which has a fundamentally pragmatic character.25 So
there is a sense in which modulation is necessary, but that is not quite the sense in
which saturation is. With saturation there is a semantic gap and an instruction
to fill the gap – both the gap and the instruction being part of the linguistic
meaning of the expression. With modulation, there need be no gap and there
is no instruction to search for some contextual filler. The expression means
something, and that meaning could go into the interpretation – so modulation
is optional – but to determine a suitable sense for complex expressions, we
need to go beyond the meaning of individual words and creatively enrich or
otherwise adjust what we are given in virtue purely of linguistic meaning. We
must go beyond linguistic meaning, without being linguistically instructed to
do so, if we are to make sense of the utterance. As Rumelhart writes, utterance
understanding involves ‘searching memory for a schema to account for the
described event’.26 I can’t resist quoting Rumelhart’s methodological comment
here:
This approach is, I believe, quite different from the ‘standard’ approach. The standard
view emphasizes the ‘bottom up’ process of constructing meaning from smaller com-
ponent meanings. Non-linguistic knowledge comes into play after the set of possible
25 See Benveniste, ‘La Forme et le Sens’, pp. 224–6, and Rumelhart, ‘Problems with Literal
Meanings’, p. 76. In The Architecture of the Language Faculty (MIT Press, 1997), pp. 47–67,
Ray Jackendoff similarly argues for what he calls ‘enriched composition’.
26 Rumelhart, ‘Problems with Literal Meanings’, p. 78.
140 Literal Meaning
meanings has been selected. My approach suggests that comprehension, like percep-
tion, should be likened to Hebb’s paleontologist (Hebb 1949) who uses his beliefs and
knowledge about dinosaurs in conjunction with the clues provided by the bone fragments
available to construct a full-fledged model of the original. In this case the words spoken
and the actions taken by the speaker are likened to the clues of the paleontologist, and the
dinosaur, to the meaning conveyed through those clues. On this view the processing is
much more ‘top down’ in that internal hypotheses are actively imposed on the observed
utterances.27
On the third type of view – the Wrong Format view (WF) – modulation is ine-
liminable for deeper reasons. It’s not just semantic composition which requires
adjustment and modulation of word meaning. Individual word meanings them-
selves are such that they could not go directly into the interpretation. They don’t
have the proper format for that. They are either too abstract and schematic, in
such a way that elaboration or fleshing out is needed to reach a determinate con-
tent;28 or they are too rich and must undergo ‘feature-cancellation’,29 or some
other screening process through which some aspects will be backgrounded and
others focused on.30 Note that there are versions of this view which take the
meaning of a word to consist both in some abstract schema in need of elabora-
tion and a large store of encyclopedic representations most of which must be
screened off as irrelevant, on any particular use.31
According to WF modulation is mandatory, hence it becomes harder to distin-
guish it from saturation. This is the point where Indexicalism and Contextualism
meet (§6.6). An indexicalist will argue that the schematicity of word meaning
invoked by (some versions of) WF just is the presence in that meaning of free
variables or parameters. So there is no modulation really, only saturation. A
contextualist will reply, as I did in §6.4, that the alleged parameters do not
behave like standard indexical variables.32
27 Ibid. I am not suggesting that Rumelhart himself advocated the Pragmatic Composition view.
I think his Contextualism was more radical than that. As the above passage shows, he held that
semantic composition is, to a large extent, a top down process. But he would presumably not
have accepted the claim that the meaning of a linguistic expression could go directly into the
interpretation, without modulation. (The same thing holds for Benveniste, whose conception of
word meaning was close to the ‘Wrong Format’ view I am about to introduce.)
28 Gustave Guillaume, Temps et Verbe (Champion, 1929), pp. 132–4; Ruhl, On Monosemy; Victorri
and Fuchs, La Polysémie.
29 See Cohen, ‘Grice’s Views about the Logical Particles’ and ‘The Semantics of Metaphor’, and
Franks, ‘Sense Generation’.
30 See for example James Pustejovsky and Branimir Boguraev, ‘A Richer Characterization of Dic-
tionary Entries: The Role of Knowledge Representations’, in Sue Atkins and Antonio Zampolli
(eds.), Computational Approaches to the Lexicon (Oxford University Press, 1994), 295–311.
31 See Langacker, Foundations of Cognitive Grammar.
32 In his contextualist manifesto, ‘Is Snow White?’, Julius Moravcsik points out that contextual
indices of the sort used to deal with indexicals have two important properties. First, they ‘simply
limit or relativize truth to a frame of application within which variations in extension and truth do
not arise’. Second, ‘for most of the standard indices we can find individual lexical items to which
these are linked’. Neither of these conditions holds in the case of modulation. The phenomenon
Contextualism: how far can we go? 141
‘is not linked to a specific sub-vocabulary of English. Practically the whole descriptive vocabu-
lary is affected.’ Moroever, ‘the various practical contexts create rather than restrict extension. It
is not as if we could survey all of the possible future extension-creating contexts that our future
and possible interactions with the world can create. Thus the intuitive idea that underlies the
practice of relativizing truth is missing in our type of case. The expanding ranges of applica-
tion and contexts cannot be described in a rigorous way, while this is possible in the cases in
which the introduction of indices is legitimate.’ (All quotations are from Julius Moravcsik, ‘Is
Snow White?’, in Paul Humphreys (ed.), Patrick Suppes: Scientific Philosopher (Kluwer, 1994),
p. 77. See also Moravcsik’s Meaning, Creativity, and the Partial Inscrutability of the Human
Mind (CSLI Publications, 1998), pp. 39–48.)
33 Friedrich Waismann, ‘Verifiability’, in Anthony Flew (ed.), Logic and Language, 1st series
(Blackwell, 1951), pp. 119–23.
142 Literal Meaning
To show that, one has only to embed the state of affairs s within a larger
situation, by providing further details about an imagined world in which s
obtains. If the world in question is sufficiently unlike our world (for example,
if the ‘cat’ talked about turns out to speak Latin, or grows to a fantastic size, or
changes into a fish), we shall be at a loss when it comes to deciding whether the
statement ‘There is a cat next door’ is actually true, with respect to that world,
even though the state of affairs we initially specified obtains. Is a world in which
there is an animal next door exactly like a cat in all respects except that it speaks
Latin a world in which there is a cat next door? We don’t know, because ‘most
of our empirical concepts are not delimited in all possible directions’.34 Or, to
take another classic example:
The notion of gold seems to be defined with absolute precision, say by the spectrum of
gold with its characteristic lines. Now what would you say if a substance was discovered
that looked like gold, satisfied all the chemical tests for gold, whilst it emitted a new sort
of radiation? ‘But such things do not happen.’ Quite so; but they might happen, and that is
enough to show that we can never exclude altogether the possibility of some unforeseen
situation arising in which we shall have to modify our definition. Try as we may, no
concept is limited in such a way that there is no room for any doubt. We introduce a
concept and limit it in some directions; for instance, we define gold in contrast to some
other metals such as alloys. This suffices for our present needs, and we do not probe any
farther. We tend to overlook the fact that there are always other directions in which the
concept has not been defined. And if we did, we could easily imagine conditions which
would necessitate new limitations. In short, it is not possible to define a concept like
gold with absolute precision, i.e. in such a way that every nook and cranny is blocked
against entry of doubt. That is what is meant by the open texture of a concept.35
Suppose that the cat and the mat are in exactly the relations depicted only they are both
floating freely in outer space, perhaps outside the Milky Way galaxy altogether . . .
Is the cat still on the mat? And was the earth’s gravitational field one of the things
depicted . . .?
What I think is correct to say as a first approximation in answer to these questions is
that the notion of the literal meaning of the sentence ‘The cat is on the mat’ does not
have a clear application, unless we make some further assumptions, in the case of cats
and mats floating freely in outer space.36
Now Waismann has an explanation for the fact that, in extraordinary situa-
tions, ‘words fail us’, as Austin puts it. Here it goes:
If I had to describe the right hand of mine which I am now holding up, I may say different
things of it: I may state its size, its shape, its colour, its tissue, the chemical compound
of its bones, its cells, and perhaps add some more particulars; but however far I go, I
shall never reach a point where my description will be completed: logically speaking,
it is always possible to extend the description by adding some detail or other. Every
description stretches, as it were, into a horizon of open possibilities: however far I go,
I shall always carry this horizon with me. (. . .) [This] has a direct bearing on the open
texture of concepts. A term is defined when the sort of situation is described in which it
is to be used. Suppose for a moment that we were able to describe situations completely
without omitting anything (as in chess), then we could produce an exhaustive list of all
the circumstances in which the term is to be used so that nothing is left in doubt; in other
words, we could construct a complete definition, i.e. a thought model which anticipates
and settles once and for all every possible question of usage. As, in fact, we can never
eliminate the possibility of some unforeseen factor emerging, we can never be quite sure
that we have included in our definition everything that should be included, and thus the
process of defining and refining an idea will go on without ever reaching a final stage.37
Waismann’s point is simple enough. A term is used in, or applies to, situations
of a certain type – situations with which we are acquainted. (Or at least: our
mastery of the term is somehow connected to our having experienced the type
of situation in question.) To define a term is to describe the type of situation in
question. The problem is that the situations in question, like Waismann’s right
hand or any aspect of empirical reality, cannot be completely described: they
possess an indefinite number of features, some of which may never have been
noticed and perhaps will never be noticed by anyone. When we describe an
empirical situation, we make certain features explicit, but an indefinite number
of other features remain implicit and constitute a sort of hidden ‘background’.
(Here I use Searle’s term, on purpose.) Now the applicability of a term to novel
situations depends on their similarity to the source situations, that is, to the
situations by association with which the term has acquired the meaning it has.
For the term to be (clearly) applicable, the target situation must be similar
to the source situations not only with respect to those features which easily
come to mind and constitute the ‘explicit’ definition of the term (the ‘tip’, to
use the iceberg metaphor), but also with respect to the hidden background. If
the two situations considerably diverge with respect to the latter, it’s unclear
whether or not the term will be applicable, even though the explicit conditions
of satisfaction are satisfied.
Other examples can be provided where the divergence between the source
situation and the target situation affects not only the hidden background but
also the explicit part (the tip of the iceberg). Thus Searle imagines that ‘Snow
is white’ is uttered as a description of the following situation:
37 Waismann, ‘Verifiability’, pp. 121–3.
144 Literal Meaning
Suppose that by some fantastic change in the course of nature the earth is hit by an
astronomical shower of radiation that affects all existing and future water molecules in
such a way that in their crystalline form they reflect a different wave length when in
sunlight from what they did prior to the radiation shower. Suppose also that the same
shower affects the human visual apparatus and its genetic basis so that snow crystals
look exactly as they did before. Physicists after the shower assure us that if we could see
snow the way we did before, it would look chartreuse but because of the change in our
retinas, which affects our observation of snow and nothing else, snow looks the same
color as ever and will continue to do so to ensuing generations . . . Would we say that
snow was still white?38
not define gold as this metal (while pointing to a piece of gold)? By explicitly
incorporating ostensive reference to the actual environment into the definition
of predicates like ‘gold’ and ‘cat’, it seems that we can overcome the alleged
unstability of truth-conditions. ‘There is a cat next door’ will be true, in the
imagined situation (where a cat-like animal next door speaks Latin), provided
the cat-like animal belongs to the same species as our cats (the cats to be found
in ordinary situations).
(Second objection) Of course, we do not know whether or not the strange
animal we imagine would be considered a cat by the scientists, just as we don’t
know whether or not a gold-like metal emitting a new sort of radiation would
count as gold. But our epistemic limitations do not prevent the word ‘gold’ (or
the word ‘cat’) from having a definite content and a definite extension: gold
is anything that is the same metal as this, and a cat is any animal of the same
species as those. Our epistemic limitations, so much emphasized by Waismann,
have no bearing on the properly semantic issue.
These objections are well-taken but they are not decisive. In particular, they
do not threaten Waismann’s conclusion (or at least, my conclusion) regarding
the unstability of truth- and application-conditions. They would threaten that
conclusion if, by appealing to ostensive definitions, we could determine a stable
content (that is, stable conditions of application) for the word thus defined. But
we cannot. For a stable content to be determined by an ostensive definition à la
Putnam, the dimension of similarity to the demonstrated exemplars must itself
be fixed.40 But what fixes the dimension of similarity? On Waismann’s picture,
words are associated with situations of use, that is all. To apply the word to or
in a novel situation, that situation must be similar to the source situations; but
we cannot survey in advance all the possible dimensions of similarity between
the source situations and possible target situations: open texture again.
According to Putnam, the predicate ‘water’ means something like: ‘same-L
as the transparent, odourless, thirst-quenching stuff to be found in lakes and
rivers in the local environment’, or more simply ‘same-L as that’ (pointing to a
sample of water). If we change the demonstrated stuff (that is, the liquid which
satisfies the stereotype of water in the local environment) we thereby change the
extension of ‘water’. ‘Water’, therefore, has an indexical component, Putnam
says. Yet there is another form of context-sensitivity in play here, which
Putnam’s ostensive definitions do not properly capture: the dimension of sim-
ilarity itself is not given, but contextually determined. In some contexts the
chemical composition of the demonstrated stuff will be relevant, in other
40 As McKay and Stern point out, ‘to determine the extension of the natural kind term, we need, in
addition to the sample, at least some indication of the breadth of the term – the respects in which
other individuals [in the extension] must be related to the sample’ (Thomas McKay and Cindy
Stern, ‘Natural Kind Terms and Standards of Membership’, in Linguistics and Philosophy 3
(1979), p. 27).
146 Literal Meaning
We now have in hand all the ingredients necessary to define the theoretical po-
sition I want to ascribe to the (radical) contextualist: the irreducibly referential
dimension of meaning; the central role of similarity in determining extension
(hence truth-conditions); and finally the context-dependence of similarity rela-
tions. With those ingredients, we can concoct an extreme view very much in
the spirit of Austin and Wittgenstein: Meaning Eliminativism.
41 A reader of Saul Kripke’s Naming and Necessity (Blackwell, 1980) might object as follows:
We are concerned only with what words mean in our language, not with what they mean
in other possible situations; now in our language water is H2 O, and ‘This is water’ is true
iff the demonstrated stuff is H2 O. This pseudo-Kripkean objection is mistaken and easy to
rebut. I am talking about our language. Our language is such that we can imagine contexts in
which it would be true to say ‘This is water’ of some stuff if the stuff in question had certain
phenomenal/functional properties, whether or not the thing in question was H2 O.
42 Noam Chomsky, New Horizons in the Study of Language and Mind (Cambridge University
Press, 2000), p. 41.
43 Putnam, ‘The Meaning of “Meaning” ’, pp. 238–9.
Contextualism: how far can we go? 147
constructed. In that framework there still is a role for the linguistic meaning of
word types: it is the input (or one of the inputs) to the construction process.
The difference between Meaning Eliminativism (ME) and WF is that, ac-
cording to ME, we don’t need linguistic meanings even to serve as input to the
construction process. The senses that are the words’ contributions to contents
are constructed, but the construction can proceed without the help of conven-
tional, context-independent word meanings.
Note that, according to a trivial extension of WF, the linguistic meaning of a
word is not merely the input to the process of semantic modulation: it is also the
output of a process of induction through which the child, or anyone learning the
language, abstracts the meaning of the word from the specific senses which it
expresses, or seems to express, on the observed occasions of use. It is a truism
that the child or language learner starts not with pre-formatted linguistic mean-
ings, but with actual uses of words and the contextualized senses that words
assume on such uses. So both contextualized senses and context-independent
linguistic meanings are input, and both are output, in some construction pro-
cess. The linguistic meaning of a word type is the output of an abstraction
process; that process takes as input the contextualized senses used as evidence
by the language learner. On the other hand, the linguistic meaning of a word
type also serves as input to the modulation process which yields as output the
contextualized sense of the word on a particular occasion of use (figure 9.1).
ME purports to simplify WF by suppressing the intermediary step (linguistic
meaning) and computing directly the contextual sense which an expression as-
sumes on a particular occasion of use on the basis of the contextual senses which
that expression had on previous occasions of use – without ever abstracting,
or needing to abstract, ‘the’ linguistic meaning of the expression type.44 This
amounts to merging the two construction processes: the abstraction of meaning
from use, and the modulation of meaning in use (figure 9.2). According to ME,
there is a single process of abstraction-modulation which takes as input previ-
ous uses of the expression and yields as output the contextual sense assumed
by the expression on the current use.
On the resulting picture, words are not primitively associated with abstract
‘conditions of application’, constituting their conventional meaning (as on the
44 For a detailed psychological model supporting ME, see Douglas Hintzman, ‘“Schema Abstrac-
tion” in a Multiple-Trace Memory Model’, in Psychological Review 93 (1986), 411–28, and
‘Judgments of Frequency and Recognition Memory in a Multiple-Trace Memory Model’, in
Psychological Review 95 (1988), 528–51.
148 Literal Meaning
MEANING ELIMINATIVISM:
of application does not end there. Even once the target-situation is fixed, the
relevant dimensions for evaluating the similarity between that situation and
the source-situations remain underdetermined: those dimensions will vary as a
function of the subject of conversation, the concerns of the speech participants,
and so on.
One particularly important factor in the contextual variation is the relevant
‘contrast set’. As Tversky has pointed out, judgments of similarity are very much
affected by variations along that dimension.45 If we ask which country, Sweden
or Hungary, most resembles Austria (without specifying the relevant dimension
of similarity), the answer will depend on the set of countries considered. If
that set includes not just Sweden, Hungary and Austria but also Poland, then
Sweden will be judged more like Austria than Hungary; but if the last of the
four countries considered is Norway and not Poland, then it is Hungary which
will be judged more like Austria than Sweden. The explanation for that fact
is simple. Poland and Hungary have certain salient geopolitical features in
common which can serve as basis for the classification: Hungary and Poland
are then put together and opposed to Austria and Sweden. If we replace Poland
by Norway in the contrast set a new principle of classification emerges, based
on the salient features shared by Norway and Sweden: in this new classification
Hungary and Austria are put together. Tversky concludes that judgments of
similarity appeal to features having a high ‘diagnostic value’ (or classificatory
significance), and that the diagnostic value of features itself depends on the
available contrast set.
So the set of similarity features on which sense depends itself depends upon
the relevant contrast set, and the relevant contrast set depends upon the current
interests of the conversational participants. It follows that one can, by simply
shifting the background interests ascribed to the conversational participants,
change the truth-conditions of a given utterance, even though the facts (including
the target-situation) don’t change, and the semantic values of indexicals remain
fixed. Charles Travis has produced dozens of examples of this phenomenon of
truth-conditional shiftiness over the last thirty years, and his examples often
involve manipulating the relevant contrast-set.46
trouble distinguishing ducks from geese. Then what Fred said may well be true. It would also be
true had Fred said what he did in pointing out the fact that all the other ducks were poor copies
(perhaps on the order of Donald Duck).’ (Saying and Understanding, p. 51).
47 See Tversky, ‘Features of similarity’, p. 342: ‘The feature “real” has no diagnostic value in the
set of actual animals since it is shared by all actual animals and hence cannot be used to classify
them. This feature, however, acquires considerable diagnostic value if the object set is extended
to include legendary animals, such as a centaur, a mermaid or a phoenix.’
48 It is easy to imagine a context with respect to which it would be definitely true to say ‘the
cat is on the mat’ of the gravitationless situation described by Searle: ‘For example, as we are
strapped in the seats of our spaceship in outer space we see a series of cat-mat pairs floating past
Contextualism: how far can we go? 151
9.7 Conclusion
Meaning Eliminativism gets rid of abstract meanings for types, in favour of
particular uses. The contextualized sense carried by the word on a particu-
lar use depends upon similarity relations between that use of the word (as
applied to the ‘target-situation’) and past uses of the same word (as applied
to the ‘source-situations’). The features of similarity that are perceived are
so context-dependent (and in particular, contrast-dependent) that the truth-
conditional sense of the word is susceptible to vary, and to vary more or less
indefinitely, even after we have fixed (i) the collection of past situations of use,
(ii) the target-situation in which or to which the word is currently applied, and
(iii) the reference of indexicals and the semantic values of all expressions in
need of saturation.
Meaning Eliminativism is, as I said, the most extreme form of Contextualism.
It is probably too extreme, but what strikes me most in that framework is
its surprising viability. The knock-down objections such an extreme view is
supposed to raise (and does raise in the classroom) do not sound so devastating
when explicitly stated. To close this chapter, let me briefly consider two standard
objections to the elimination of linguistic meaning.
The first objection (which I may have heard about fifty times) runs like this.
Granted, the senses expressed by words are affected by contextual factors;
they are ‘modulated’. But we must treat modulation effects as arising from the
interaction of context with context-independent linguistic meanings. If, instead,
we go eliminativist and say that the context does everything, while the words
themselves contribute nothing, then it does not matter any more whether the
word one utters is ‘red’ or ‘rectangle’. But it makes a big difference which
word we actually utter. This shows that Meaning Eliminativism goes too far: it
throws out the baby with the bathwater.
our window. Oddly, they come in only two attitudes. From our point of view they are either as
depicted in figure 1 or as would be depicted if figure 1 were upside down. “Which is it now?”,
I ask. “The cat is on the mat”, you answer. Have you not said exactly and literally what you
meant?’ (Searle, ‘Literal Meaning’, p. 212.)
152 Literal Meaning
This objection presupposes that, for the eliminativist, ‘the words themselves
contribute nothing’. But that is not true. There is something which words do
contribute, and which is sufficient to account for the difference it makes when we
substitute a word for another one. That thing is not a ‘meaning’ in the traditional
sense: it is what I called the ‘semantic potential’ of words. ‘Red’ and ‘rectangle’
have very different semantic potentials. The semantic potential of a word is
the collection of past uses on the basis of which similarities can be established
between the source situations (that is, the situations such uses concerned) and the
target situation (the situation of current concern). So there is, indeed, interaction
between the ‘context’ and what words themselves contribute. Context does not
do everything, even on the most extreme form of Contextualism.
According to the second objection, there is actually more stability in se-
mantic content than ME suggests. Not only do words have relatively stable
conditions of application, across uses; there is also interpersonal stability: lan-
guage users converge in their judgments regarding conditions of application or
truth-conditions. How will that stability be accounted for?
With regard to the first part of the objection the contextualist pleads guilty:
the unstability and shiftiness of truth-conditional content across uses he takes
to be an empirical fact, sufficiently established by the wealth of examples and
thought-experiments provided by the authors I have mentioned. Still, accord-
ing to the objection, the contextualist goes too far in that direction. But the
contextualist will deny that that is so. The degree of unstability achieved will
depend upon details of the model, and can be varied so as to fit experimental
data. (See Hintzman’s papers for suggestions.) As for the second part of the
objection, the contextualist will be happy to grant the interpersonal stability of
truth-conditional content. That stability can be accounted for on psychological
rather than linguistic grounds. Thus Bolinger notes that
speakers not only share the same code but also share the ability to see the same re-
semblances between what their code already designates and what they would like it to
designate, and so to make the old forms reach out to new meanings. That is how language
breaks free of its rigidity.49
In any event, there are all sorts of versions of the contextualist position
which are intermediate between WF and ME. Like Rumelhart and most lexi-
cal semanticists, we may consider that words are associated not so much with
(a collection of) particular, episodic situations of use, but with more abstract
schemata corresponding to types of situations. Insofar as there still is a referen-
tial or quasi-referential dimension in virtue of which, through these schemata,
words point to actual situations of use, the basic mechanisms posited by ME
remain in play. Or we may favour a model in which both episodic traces of
situations of use and abstract schemata are stored and exploited in contextually
determining truth-conditional content. Whichever option we take the important
contextualist idea should not be misconstrued. How much abstraction takes
place is not the issue, for the contextualist. The important idea – which forms
the core of Austin’s theory of truth – is twofold: first, words are associated with
worldly situations (at whatever level of abstractness and fine-grainedness), that
is, entities at the level of reference rather than the level of sense; second, the
content (sense) of words depends upon context-sensitive similarities between
those situations and the target-situation in which or to which the word is applied.
Conclusion
At the beginning of chapter 6, I said that the debate between Literalism and
Contextualism, which stood at the forefront of attention in the middle of the
twentieth century, is widely, and wrongly, believed to have been settled in
favour of (an attenuated version of) Literalism. My aim, in this book, has been
to revive that debate, and to show that Contextualism is still a live option.
By ‘Contextualism’ I mean the view according to which it is speech acts, not
sentences, which have a determinate content and are truth-evaluable: sentences
themselves express a determinate content only in the context of a speech act.
In this concluding chapter I will summarize the discussion and deal with a few
residual issues.
1 Peter Geach, ‘Assertion’, in Logic Matters (Blackwell, 1972), pp. 254–5. By ‘proposition’ Geach
means ‘a form of words in which something is propounded, put forward for consideration’
(p. 255).
154
Conclusion 155
order for the complex sentence in which it occurs itself to express a determinate,
truth-evaluable content:
Oxford-trained philosophers often say nowadays that a sentence can have a truth value
assigned to it only in that it is ‘used to make a statement’ in a given context. If this
were literally true, then a truth-functional account of ‘p vel q’ or of ‘p aut q’ would
be impossible: for the disjunct clauses represented by ‘p’ and ‘q’ would not be being
‘used to make statements’ in a context in which only the disjunction was asserted, and
would thus not have any truth values for the truth value of the whole proposition to be
a function of.2
But this objection misses its target. What matters, from the contextualist point
of view, is not whether or not a sentence is asserted. Sentences are said to carry
determinate contents only insofar as they are used to make a statement (or to
perform another speech act). The unasserted clauses that occur in condition-
als or disjunctions satisfy this constraint no less than the complex sentences in
which they occur: the speech act which the complex sentence is used to perform
is sufficient to endow the constituent sentences (as much as the complex sen-
tence itself) with a determinate content. It is not necessary that the constituent
sentences be used to perform a speech act of their own. Thus I agree with Aaron
Snyder’s rebuttal of Geach’s argument:
Geach claims that, in a context in which only the disjunction, ‘p vel q’ was asserted, the
disjunct clauses represented by ‘p’ and ‘q’ would not be being used to make statements.
However, they certainly would be being used to make the statement that p vel q. It might
be that Geach means that they would not be being used to make the statements that p and
that q, respectively. But if this is what he means, his argument is an ignoratio elenchi, for
it counts against only such specific theories as would hold that the sentence represented
by ‘p’ has truth value only in that it is used to make the statement that p.3
Note that the speech acts (or thought acts) in the context of which sentences
are said to carry determinate contents need not be construed too restrictively.
The list will include supposing that p and making as if to say that p, as well as
asserting that p.
I turn to the Gricean argument, the influence of which was decisive in the
downfall of Contextualism. That argument actually is a counterargument: it
aims at undermining the standard contextualist argument, which we may call
‘the argument from the contextual variability of truth-conditions’. The contex-
tualist argument runs as follows: in an indefinite number of cases, one and the
same sentence can be used to make different statements, with different truth-
conditions, depending on the context; this shows that the sentence itself does not
carry definite truth-conditions. For example, in Introduction to Logical Theory,
2 Ibid., p. 258. Geach uses the Latin words ‘vel’ and ‘aut’ as connectives corresponding to the
inclusive (‘vel’) and exclusive (‘aut’) readings of ‘or’.
3 Aaron Snyder, ‘On What Has Truth-Value’, in Philosophical Studies 23 (1972), p. 131.
156 Literal Meaning
By virtue of Modified Occam’s Razor, the analyst who observes that a sentence
has two different interpretations when uttered in different contexts must refrain
from considering that this intuitive difference in interpretation reflects a differ-
ence in linguistic meaning, that is, a semantic ambiguity. She must, if possible,
ascribe this difference to a property of the context of utterance rather than to an
ambiguity in the sentence itself.
Together, Modified Occam’s Razor and the Parallelism Principle entail that
the analyst must not only refrain from considering that a contextual difference
in interpretation reflects a difference in linguistic meaning, but also refrain
from considering that it reflects a difference in propositional content, in ‘what
is said’ by the sentence when it is uttered in this or that context. For suppose we
take the difference in interpretation to be a difference in propositional content.
Then, by virtue of the Parallelism Principle, this difference in content is to be
explained in terms of a difference in linguistic meaning (semantic ambiguity);
but positing a semantic ambiguity is precisely what Modified Occam’s Razor
says the analyst should refrain from doing. It follows that the analyst must, if
possible, explain the contextual difference in interpretation by a property of the
context of utterance, while maintaining that the sentence itself is not ambiguous
and that ‘what is strictly and literally said’ (the propositional content) does not
change from one context to the next.
A general solution along these lines has been sketched by Grice, based on
the notion of conversational implicature. A conversational implicature is some-
thing which is communicated by an utterance and therefore belongs to its overall
interpretation, but which belongs neither to the linguistic meaning of the sen-
tence uttered nor to what is said by the utterance of this sentence; it is an
aspect of interpretation that is external to what is said. (Indeed, working out
the implicatures of an utterance presupposes identifying what this utterance
‘says’.) Modified Occam’s Razor and the Parallelism Principle lead one to
favour an analysis that accounts for a contextual difference in interpretation in
terms of conversational implicature, over an account in terms of a variation in
propositional content. This is, basically, the Gricean argument against Contex-
tualism. In the case of ‘and’, the argument proceeds as follows:
(i) Instead of saying, with Strawson, that there is a semantic difference between
‘p ∧ q’ and ‘p and q’, we may consider that there is no such difference
and that the truth-conditions of ‘p and q’ are actually the same as those of
‘p ∧ q’. For example, we may consider the temporal implication in ‘They got
married and had a child’ as a conversational implicature, external to what is said,
rather than considering it as part of the truth-conditions of the utterance in a certain
type of context. In this way, we are able to maintain that the truth-conditions of ‘p
and q’ are determined by the truth-table for ‘∧’, independently of the context of
utterance.
(ii) By virtue of Modified Occam’s Razor and the Parallelism Principle, the account
in terms of conversational implicature is preferable to the account in terms of a
truth-conditional difference between ‘p ∧ q’ and ‘p and q’.
6 In actual practice, the Parallelism Principle is not explicitly stated as a premiss of the Gricean
argument. However, Modified Occam’s Razor (or an equivalent principle of economy) is used
to support the conclusion that an account in terms of implicature is preferable to an account
in terms of truth-conditional content, and I believe that Modified Occam’s Razor can be used
to that effect only if one accepts the Parallelism Principle. So the latter is actually presupposed
by the Gricean argument. (In ‘Contextualism and Anti-Contextualism’, I gave two examples
of actual uses of the Gricean argument. The first example is its use by Grice himself, against
Strawson’s treatment of ‘and’. The second example is a classical use of the Gricean argument –
by Kripke and many others – against Donnellan’s view of the referential/attributive distinction.
The Gricean argument against Donnellan is extensively discussed in my ‘Referential/Attributive:
A Contextualist Proposal’, in Philosophical Studies 56 (1989), 217–49, and in Direct Reference,
Conclusion 159
pp. 284–8. See also Anne Bezuidenhout, ‘Pragmatics, Semantic Underdetermination, and the
Referential/Attributive Distinction’, in Mind 106 (1997), 375–409.)
7 As Barwise and Perry write, ‘even if I am fully convinced that I am Napoleon, my use of “I”
designates me, not him. Similarly, I may be fully convinced that it is 1789, but it does not make
my use of “now” about a time in 1789’ (Situations and Attitudes, p. 148).
160 Literal Meaning
sentence is a standard (that is, an already familiar type of) indexical sentence.
This simply begs the question and hinders research.
I conclude that, even if one is attracted to literalist conclusions, one should
not beg the question and take those conclusions for granted. One should at least
adhere to a weak form of Contextualism, which I have called ‘Methodological
Contextualism’.8 Methodological Contextualism says that there is, in principle,
a difference between the linguistic meaning of the sentence and what is said
by an utterance of the sentence, and a correlative difference between the lin-
guistic meaning (or semantic potential) of an expression and the contribution
the expression makes to the proposition expressed by the sentence where it
occurs. Any expression may, upon analysis, turn out to have a context-sensitive
semantics, that is, be such that its truth-conditional contribution varies across
contexts. Whether or not that is the case, in a particular instance, is a matter for
empirical analysis, not for a priori stipulation.
In semantics, over the past twenty years, those lessons have been drawn.
The distinction between linguistic meaning and propositional content has been
generalized, and a number of new sources of context-sensitivity have been dis-
covered in natural language constructions. The indexicalist position defended
by Jason Stanley is typical in this regard. In his framework indexical variables
can be found all over the place; and the role of pragmatics and speaker’s mean-
ing in contextually assigning values to those variables is fully acknowledged.
Does this mean that Literalism has been defeated? No. Literalism survives, in
a weaker form, through the minimalist principle, which Stanley formulates as
follows:
All effects of extra-linguistic context on the truth-conditions of an assertion are traceable
to elements in the actual syntactic structure of the sentence uttered.9
what is said must be available to the language users.21 But he construes availabil-
ity dispositionally rather than occurrently. Carpintero therefore distinguishes
three positions. The ‘Wittgensteinians’, as he calls the contextualists, hold that
what is said must be available in the strong sense, that is, can easily become ob-
jects of conscious, explicit, occurrent judgments (= ‘CEO’ judgments) made
by competent users of the language. The ‘Chomskians’ reject the Availabil-
ity Principle and appeal only to tacit knowledge of the sub-personal variety.
Carpintero himself advocates an intermediate, ‘Gricean’ position, according
to which semantic knowledge (including knowledge of what is said) ‘is con-
stituted by personal-level psychological states, like those easily accessible in
the form of CEO-thoughts, and unlike the Chomskian states of tacit knowl-
edge; but, like the latter, it is only available after reflection heuristically of
a scientific character, taking as data intuitions about the evaluation of utter-
ances relative to several circumstances’.22 This position, Carpintero says, ‘tries
to navigate a middle course between the Scylla of the subpersonal view, and
the Charibdis of the surface-psychological view, defended by Recanati and
others’.23
I myself have given an analogous interpretation of ‘availability’ in chapter 3,
when responding to objections from relevance theorists. The inference which
gives rise to implicatures is available to interpreters, I said, but this availability
need not be cashed out as involving on the part of the subject a CEO-inference.
A tacit inference is sufficient, provided it is of the ‘personal’ variety, that is,
provided the subject herself has the reflective capacities for making the inference
explicit.24 If we generalize this dispositional interpretation of availability, the
Availability Principle can no longer be used to rule out the minimal proposition,
on the grounds that it is not ‘available’ to the language users. On the contrary,
it turns out that the pragmatic, Gricean conception of saying which I argued for
in chapter 1 is compatible with the standard, minimalist conception of what is
strictly and literally said.
In arguing against Minimalism in this book, I have taken for granted the
strong availability requirement – the requirement that what is said be ‘CEO-
available’. But that requirement is too strong, according to Carpintero. Weak
(that is, dispositional) availability is sufficient:
I share with Recanati and others the view that some form of ‘first-personal availability’ is
required. However, the one attainable through the process of ordinary scientific-semantic
theorizing is in my view quite enough. It also involves psychological processes, although
not the sort of psychological process on which Recanati et al. focus: it requires us to
21 ‘I should say straight away that I sympathize to a certain extent with the attraction to assumptions,
such as Recanati’s Availability Principle. The Gricean picture of language use emphasizes that it
is a rational process. Merely “sub-personal-level” states cannot account for this’ (ibid., p. 121).
22 Carpintero, ‘Gricean Rational Reconstructions’, p. 123.
23 Ibid., p. 121. 24 See above, p. 50.
164 Literal Meaning
speaker is abiding by the conversational maxims, together with this first blush blatant violation
of the first maxim of quality, then takes the audience (as indeed intended by the speaker) to
the non-literal (. . .) one. ‘Bill is not himself today’ (which will be interpreted as non-literally
meaning that today Bill lacks some of the attributes that characterize him) is a good example of
this.’ (p. 119).
But this is precisely what I deny when I insist that primary pragmatic processes take place
locally and are not ‘inferential’. See above, §2.2 (pp. 27–9). For a review of the psychological
literature on how non-literal meaning is actually processed, see Gibbs, The Poetics of Mind.
28 Carpintero, ‘Gricean Rational Reconstructions’, p. 124.
Bibliography
166
Bibliography 167
Benveniste, Emile, ‘La Forme et le Sens dans le Langage’, in his Problèmes de Linguis-
tique Générale II (Gallimard, 1974), 215–38.
Berg, Jonathan, ‘In Defense of Direct Belief: Substitutivity, Availability, and Iterability’,
in Lingua e Stile 33 (1998), 461–70.
Bezuidenhout, Anne, ‘Pragmatics, Semantic Underdetermination, and the Referential/
Attributive Distinction’, in Mind 106 (1997), 375–409.
Bianchi, Claudia, La Dipendenza Contestuale: Per una Teoria Pragmatica del
Significato (Edizioni Scientifiche Italiane, 2001).
Blackburn, Patrick, ‘Tense, Temporal Reference and Tense Logic’, in Journal of
Semantics 11 (1994), 83–101.
Blackburn, Patrick, de Rijke, Maarten and Venema, Yde, Modal Logic (Cambridge
University Press, 2001).
Blakemore, Diane, Understanding Utterances: An Introduction to Pragmatics
(Blackwell, 1992).
Bolinger, Dwight, Aspects of Language (Harcourt, Brace and World, 1968).
Borg, Emma, ‘Saying What You Mean: Unarticulated Constituents and Communication’
(typescript).
Braun, David, ‘Demonstratives and Their Linguistic Meaning’, in Noûs 30 (1996),
145–73.
Burge, Tyler, ‘Content Preservation’, in Philosophical Review 102 (1993), 457–88.
Campbell, Robin, ‘Language Acquisition, Psychological Dualism and the Definition of
Pragmatics’, in Herman Parret, Marina Sbisà and Jeff Verschueren (eds.), Possibil-
ities and Limitations of Pragmatics (Benjamins, 1981), 93–103.
Cappelen, Herman and Lepore, Ernie, ‘On an Alleged Connection Between
Indirect Speech and the Theory of Meaning’, in Mind and Language 12 (1997),
278–96.
‘Radical and Moderate Pragmatics: Does Meaning Determine Truth-Conditions?’
(typescript).
Caramazza, Alfonso and Grober, Ellen, ‘Polysemy and the Structure of the Subjective
Lexicon’, in Clea Rameh (ed.), Semantics: Theory and Application (Georgetown
University Press, 1976), 181–207.
Carston, Robyn, ‘Implicature, Explicature, and Truth-Theoretic Semantics’, in Ruth
Kempson (ed.), Mental Representations: The Interface between Language and
Reality (Cambridge University Press, 1988), 155–81.
‘Enrichment and Loosening: Complementary Processes in Deriving the Proposition
Expressed?’, in Linguistische Berichte 8 (1997), 103–27.
‘Explicature and Semantics’, in UCL Working Papers in Linguistics 12 (2000), 1–44.
‘Linguistic Meaning, Communicated Meaning, and Cognitive Pragmatics’, in Mind
and Language 17 (2002), 127–48.
Thoughts and Utterances: The Pragmatics of Explicit Communication (Blackwell,
2002).
‘Semantics and Conversational Implicature’, forthcoming in the Proceedings of the
Genoa conference WOC 2002: The Semantics/Pragmatics Distinction, to be pub-
lished by CSLI Publications.
‘Report on typescript by François Recanati called Literal Meaning’ (typescript).
Chierchia, Gennaro, ‘Scalar Implicatures, Polarity Phenomena, and the Syntax/
Pragmatics Interface’ (forthcoming).
168 Bibliography
Moravcsik, Julius, ‘Is Snow White?’, in Paul Humphreys (ed.), Patrick Suppes: Scientific
Philosopher (Kluwer, 1994), 71–85.
Meaning, Creativity, and the Partial Inscrutability of the Human Mind (CSLI
Publications, 1998).
Murphy, Gregory ‘Noun Phrase Interpretation and Conceptual Combination’, in Journal
of Memory and Language 29 (1990), 259–88.
‘The Comprehension of Complex Concepts’, in Cognitive Science 12 (1988),
529–62.
Neale, Stephen, This, That and the Other, chapter 1 (typescript).
Nunberg, Geoffrey, ‘The Non-Uniqueness of Semantic Solutions: Polysemy’, in
Linguistics and Philosophy 3 (1979), 143–84.
‘Indexicality and Deixis’, in Linguistics and Philosophy 16 (1993), 1–43.
‘Transfers of Meaning’, in Journal of Semantics 12 (1995), 109–32.
Nunberg, Geoff and Zaenen, Annie, ‘Systematic Polysemy in Lexicology and Lexicog-
raphy’, in Hannu Tommola, Krista Varantola, Tarja Tolonen and Jürgen Schopp
(eds.), Proceedings of Euralex 2 (University of Tampere, 1992), Part II, 387–98.
Partee, Barbara, ‘Some Structural Analogies Between Tenses and Pronouns in English’,
in Journal of Philosophy 70 (1973), 601–9.
‘Compositionality’, in Fred Landman and Frank Veltman (eds.), Varieties of Formal
Semantics (Foris, 1984), 281–312.
‘Binding Implicit Variables in Quantified Contexts’, in CLS 25 (1989), 342–65.
Pelczar, Michael, ‘Wittgensteinian Semantics’, in Noûs 34 (2000), 483–516.
Perry, John, ‘Thought Without Representation’, reprinted (with a postscript) in his col-
lection The Problem of the Essential Indexical and Other Essays (Oxford University
Press, 1993), 205–25.
The Problem of the Essential Indexical and Other Essays (Oxford University Press,
1993).
Reference and Reflexivity (CSLI Publications, 2001).
Prandi, Michele, Grammaire Philosophique des Tropes (Editions de Minuit, 1992).
Prior, Arthur, Past, Present, and Future (Clarendon Press, 1967).
‘Egocentric Logic’, in Arthur Prior and Kit Fine, Worlds, Times and Selves
(Duckworth, 1977), 28–45.
‘Now’, in Noûs 1 (1968), 101–19.
Pustejovsky, James, ‘The Generative Lexicon’, in Computational Linguistics 17 (1991),
409–41.
The Generative Lexicon (MIT Press, 1995).
Pustejovsky, James and Boguraev, Branimir, ‘A Richer Characterization of Dictionary
Entries: The Role of Knowledge Representations’, in Sue Atkins and Antonio
Zampolli (eds.), Computational Approaches to the Lexicon (Oxford University
Press, 1994), 295–311.
Putnam, Hilary, ‘The Meaning of “Meaning” ’, in his Philosophical Papers, vol. 2: Mind,
Language and Reality (Cambridge University Press, 1975), 215–71.
Quine, Willard van Orman, ‘Variables Explained Away’, in his Selected Logic Papers
(Harvard University Press, 1995), 227–35.
Recanati, François, La Transparence et l’Enonciation (Editions du Seuil, 1979).
‘Insinuation et Sous-Entendu’, in Communications 30 (1979), 95–106.
‘On Defining Communicative Intentions’, in Mind and Language 1 (1986), 213–42.
172 Bibliography
abstraction, 2, 59–61, 64, 147, 153, 164 Blackburn, P., 116, 118
accessibility, 30–4, 37, 63 Blakemore, D., 38
action (explanation of), 54–5 Blending, 79
activation see accessibility Boguraev, B., 140
ad hoc concepts, 25–6, 46–7, 87 Bolinger, D., 131, 152
adjustment see modulation, mutual Borg, E., 10, 92–3
adjustment bottom-up vs top-down see
Akman, V., 46 mandatory/optional
Almog, J., 56 Boucher, J., 25
ambiguity, 156–7, 159; see also Bouquet, P., 46
disambiguation, polysemy Brand, M., 108
Ariel, M., 30 Braun, D., 57
Atkins, B.S., 31, 94, 135, 140 Brentano, F., 16
Atlas, J., 134 Brewer, W., 36
Austin, J., 1–3, 84, 115, 141, 143, 146, 149, bridging, 13, 23–4, 44–5, 48
153 Brockman, J., 40
Austinian proposition, 122–4, 128–9 Bruce, B., 36
availability, 11–17, 20–3, 42, 46, 50–1, 53, Burge, T., 38
73–5, 77–81, 161–5; availability
constraint/condition, 17, 20, 42–4, 50, 53; Campbell, R., 12
Availability Principle, 14–15, 162–3 cancellability, 137, 162
Cappelen, H., 15, 92–3
Bach, K., 8–9, 15, 19, 32–4, 40, 52–6, 59, 74, Caramazza, A., 135
77, 108, 129, 161–2 Carnap, R., 1–2
background, 3, 6, 39, 44, 54–6, 81, 89–92, 96, Carpintero, M. see Garcia-Carpintero
108–9, 142–4, 149–50 Carruthers, P., 25
Bar-Hillel, Y., 26 Carston, R., 13, 24–6, 38–40, 44–9, 77–9,
Barker, C., 13 88–9, 95, 97
Barsalou, L., 25, 30 Casati, R., 28
Bartsch, R., 26, 29, 77 CEO states and processes, 49–50, 162–5
Bates, M., 135 character, 56–8, 64, 69, 159
Barwise, J., 122–3, 127, 159 Chierchia, G., 17
Beaney, M., 122 Chomsky, N., 108, 146
Benveniste, E., 135, 139–40 circumstance, 107, 109; of evaluation, 25, 99,
Berg, J., 50, 53 115–30; sub-sentential, 125–7, 137
Bezuidenhout, A., 126, 159 Clark, H., 24, 44–5, 77–8, 134
Bianchi, C., 95, 149 coercion, 36–7
Bierwisch, M., 44 Cohen, J., 26, 89, 131–3, 135–7, 140
Billman, D., 30 Cole, P., 29
Binding Argument, 104, 110–12 common ground, 48
Binding Criterion, 102–7, 111 comparison classes see parametric predicates
Binding Fallacy, 98, 109–13 completion, 52–3, 74
175
176 Index